xref: /freebsd/contrib/llvm-project/clang/lib/Format/UnwrappedLineParser.cpp (revision 3750ccefb8629a08890bfbae894dd6bc6a7483b4)
1 //===--- UnwrappedLineParser.cpp - Format C++ code ------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 ///
9 /// \file
10 /// This file contains the implementation of the UnwrappedLineParser,
11 /// which turns a stream of tokens into UnwrappedLines.
12 ///
13 //===----------------------------------------------------------------------===//
14 
15 #include "UnwrappedLineParser.h"
16 #include "FormatToken.h"
17 #include "FormatTokenLexer.h"
18 #include "FormatTokenSource.h"
19 #include "Macros.h"
20 #include "TokenAnnotator.h"
21 #include "clang/Basic/TokenKinds.h"
22 #include "llvm/ADT/STLExtras.h"
23 #include "llvm/ADT/StringRef.h"
24 #include "llvm/Support/Debug.h"
25 #include "llvm/Support/raw_os_ostream.h"
26 #include "llvm/Support/raw_ostream.h"
27 
28 #include <algorithm>
29 #include <utility>
30 
31 #define DEBUG_TYPE "format-parser"
32 
33 namespace clang {
34 namespace format {
35 
36 namespace {
37 
38 void printLine(llvm::raw_ostream &OS, const UnwrappedLine &Line,
39                StringRef Prefix = "", bool PrintText = false) {
40   OS << Prefix << "Line(" << Line.Level << ", FSC=" << Line.FirstStartColumn
41      << ")" << (Line.InPPDirective ? " MACRO" : "") << ": ";
42   bool NewLine = false;
43   for (std::list<UnwrappedLineNode>::const_iterator I = Line.Tokens.begin(),
44                                                     E = Line.Tokens.end();
45        I != E; ++I) {
46     if (NewLine) {
47       OS << Prefix;
48       NewLine = false;
49     }
50     OS << I->Tok->Tok.getName() << "["
51        << "T=" << (unsigned)I->Tok->getType()
52        << ", OC=" << I->Tok->OriginalColumn << ", \"" << I->Tok->TokenText
53        << "\"] ";
54     for (SmallVectorImpl<UnwrappedLine>::const_iterator
55              CI = I->Children.begin(),
56              CE = I->Children.end();
57          CI != CE; ++CI) {
58       OS << "\n";
59       printLine(OS, *CI, (Prefix + "  ").str());
60       NewLine = true;
61     }
62   }
63   if (!NewLine)
64     OS << "\n";
65 }
66 
67 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line) {
68   printLine(llvm::dbgs(), Line);
69 }
70 
71 class ScopedDeclarationState {
72 public:
73   ScopedDeclarationState(UnwrappedLine &Line, llvm::BitVector &Stack,
74                          bool MustBeDeclaration)
75       : Line(Line), Stack(Stack) {
76     Line.MustBeDeclaration = MustBeDeclaration;
77     Stack.push_back(MustBeDeclaration);
78   }
79   ~ScopedDeclarationState() {
80     Stack.pop_back();
81     if (!Stack.empty())
82       Line.MustBeDeclaration = Stack.back();
83     else
84       Line.MustBeDeclaration = true;
85   }
86 
87 private:
88   UnwrappedLine &Line;
89   llvm::BitVector &Stack;
90 };
91 
92 } // end anonymous namespace
93 
94 std::ostream &operator<<(std::ostream &Stream, const UnwrappedLine &Line) {
95   llvm::raw_os_ostream OS(Stream);
96   printLine(OS, Line);
97   return Stream;
98 }
99 
100 class ScopedLineState {
101 public:
102   ScopedLineState(UnwrappedLineParser &Parser,
103                   bool SwitchToPreprocessorLines = false)
104       : Parser(Parser), OriginalLines(Parser.CurrentLines) {
105     if (SwitchToPreprocessorLines)
106       Parser.CurrentLines = &Parser.PreprocessorDirectives;
107     else if (!Parser.Line->Tokens.empty())
108       Parser.CurrentLines = &Parser.Line->Tokens.back().Children;
109     PreBlockLine = std::move(Parser.Line);
110     Parser.Line = std::make_unique<UnwrappedLine>();
111     Parser.Line->Level = PreBlockLine->Level;
112     Parser.Line->PPLevel = PreBlockLine->PPLevel;
113     Parser.Line->InPPDirective = PreBlockLine->InPPDirective;
114     Parser.Line->InMacroBody = PreBlockLine->InMacroBody;
115     Parser.Line->UnbracedBodyLevel = PreBlockLine->UnbracedBodyLevel;
116   }
117 
118   ~ScopedLineState() {
119     if (!Parser.Line->Tokens.empty())
120       Parser.addUnwrappedLine();
121     assert(Parser.Line->Tokens.empty());
122     Parser.Line = std::move(PreBlockLine);
123     if (Parser.CurrentLines == &Parser.PreprocessorDirectives)
124       Parser.MustBreakBeforeNextToken = true;
125     Parser.CurrentLines = OriginalLines;
126   }
127 
128 private:
129   UnwrappedLineParser &Parser;
130 
131   std::unique_ptr<UnwrappedLine> PreBlockLine;
132   SmallVectorImpl<UnwrappedLine> *OriginalLines;
133 };
134 
135 class CompoundStatementIndenter {
136 public:
137   CompoundStatementIndenter(UnwrappedLineParser *Parser,
138                             const FormatStyle &Style, unsigned &LineLevel)
139       : CompoundStatementIndenter(Parser, LineLevel,
140                                   Style.BraceWrapping.AfterControlStatement,
141                                   Style.BraceWrapping.IndentBraces) {}
142   CompoundStatementIndenter(UnwrappedLineParser *Parser, unsigned &LineLevel,
143                             bool WrapBrace, bool IndentBrace)
144       : LineLevel(LineLevel), OldLineLevel(LineLevel) {
145     if (WrapBrace)
146       Parser->addUnwrappedLine();
147     if (IndentBrace)
148       ++LineLevel;
149   }
150   ~CompoundStatementIndenter() { LineLevel = OldLineLevel; }
151 
152 private:
153   unsigned &LineLevel;
154   unsigned OldLineLevel;
155 };
156 
157 UnwrappedLineParser::UnwrappedLineParser(
158     SourceManager &SourceMgr, const FormatStyle &Style,
159     const AdditionalKeywords &Keywords, unsigned FirstStartColumn,
160     ArrayRef<FormatToken *> Tokens, UnwrappedLineConsumer &Callback,
161     llvm::SpecificBumpPtrAllocator<FormatToken> &Allocator,
162     IdentifierTable &IdentTable)
163     : Line(new UnwrappedLine), MustBreakBeforeNextToken(false),
164       CurrentLines(&Lines), Style(Style), IsCpp(Style.isCpp()),
165       LangOpts(getFormattingLangOpts(Style)), Keywords(Keywords),
166       CommentPragmasRegex(Style.CommentPragmas), Tokens(nullptr),
167       Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1),
168       IncludeGuard(Style.IndentPPDirectives == FormatStyle::PPDIS_None
169                        ? IG_Rejected
170                        : IG_Inited),
171       IncludeGuardToken(nullptr), FirstStartColumn(FirstStartColumn),
172       Macros(Style.Macros, SourceMgr, Style, Allocator, IdentTable) {
173   assert(IsCpp == LangOpts.CXXOperatorNames);
174 }
175 
176 void UnwrappedLineParser::reset() {
177   PPBranchLevel = -1;
178   IncludeGuard = Style.IndentPPDirectives == FormatStyle::PPDIS_None
179                      ? IG_Rejected
180                      : IG_Inited;
181   IncludeGuardToken = nullptr;
182   Line.reset(new UnwrappedLine);
183   CommentsBeforeNextToken.clear();
184   FormatTok = nullptr;
185   MustBreakBeforeNextToken = false;
186   IsDecltypeAutoFunction = false;
187   PreprocessorDirectives.clear();
188   CurrentLines = &Lines;
189   DeclarationScopeStack.clear();
190   NestedTooDeep.clear();
191   NestedLambdas.clear();
192   PPStack.clear();
193   Line->FirstStartColumn = FirstStartColumn;
194 
195   if (!Unexpanded.empty())
196     for (FormatToken *Token : AllTokens)
197       Token->MacroCtx.reset();
198   CurrentExpandedLines.clear();
199   ExpandedLines.clear();
200   Unexpanded.clear();
201   InExpansion = false;
202   Reconstruct.reset();
203 }
204 
205 void UnwrappedLineParser::parse() {
206   IndexedTokenSource TokenSource(AllTokens);
207   Line->FirstStartColumn = FirstStartColumn;
208   do {
209     LLVM_DEBUG(llvm::dbgs() << "----\n");
210     reset();
211     Tokens = &TokenSource;
212     TokenSource.reset();
213 
214     readToken();
215     parseFile();
216 
217     // If we found an include guard then all preprocessor directives (other than
218     // the guard) are over-indented by one.
219     if (IncludeGuard == IG_Found) {
220       for (auto &Line : Lines)
221         if (Line.InPPDirective && Line.Level > 0)
222           --Line.Level;
223     }
224 
225     // Create line with eof token.
226     assert(eof());
227     pushToken(FormatTok);
228     addUnwrappedLine();
229 
230     // In a first run, format everything with the lines containing macro calls
231     // replaced by the expansion.
232     if (!ExpandedLines.empty()) {
233       LLVM_DEBUG(llvm::dbgs() << "Expanded lines:\n");
234       for (const auto &Line : Lines) {
235         if (!Line.Tokens.empty()) {
236           auto it = ExpandedLines.find(Line.Tokens.begin()->Tok);
237           if (it != ExpandedLines.end()) {
238             for (const auto &Expanded : it->second) {
239               LLVM_DEBUG(printDebugInfo(Expanded));
240               Callback.consumeUnwrappedLine(Expanded);
241             }
242             continue;
243           }
244         }
245         LLVM_DEBUG(printDebugInfo(Line));
246         Callback.consumeUnwrappedLine(Line);
247       }
248       Callback.finishRun();
249     }
250 
251     LLVM_DEBUG(llvm::dbgs() << "Unwrapped lines:\n");
252     for (const UnwrappedLine &Line : Lines) {
253       LLVM_DEBUG(printDebugInfo(Line));
254       Callback.consumeUnwrappedLine(Line);
255     }
256     Callback.finishRun();
257     Lines.clear();
258     while (!PPLevelBranchIndex.empty() &&
259            PPLevelBranchIndex.back() + 1 >= PPLevelBranchCount.back()) {
260       PPLevelBranchIndex.resize(PPLevelBranchIndex.size() - 1);
261       PPLevelBranchCount.resize(PPLevelBranchCount.size() - 1);
262     }
263     if (!PPLevelBranchIndex.empty()) {
264       ++PPLevelBranchIndex.back();
265       assert(PPLevelBranchIndex.size() == PPLevelBranchCount.size());
266       assert(PPLevelBranchIndex.back() <= PPLevelBranchCount.back());
267     }
268   } while (!PPLevelBranchIndex.empty());
269 }
270 
271 void UnwrappedLineParser::parseFile() {
272   // The top-level context in a file always has declarations, except for pre-
273   // processor directives and JavaScript files.
274   bool MustBeDeclaration = !Line->InPPDirective && !Style.isJavaScript();
275   ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
276                                           MustBeDeclaration);
277   if (Style.Language == FormatStyle::LK_TextProto)
278     parseBracedList();
279   else
280     parseLevel();
281   // Make sure to format the remaining tokens.
282   //
283   // LK_TextProto is special since its top-level is parsed as the body of a
284   // braced list, which does not necessarily have natural line separators such
285   // as a semicolon. Comments after the last entry that have been determined to
286   // not belong to that line, as in:
287   //   key: value
288   //   // endfile comment
289   // do not have a chance to be put on a line of their own until this point.
290   // Here we add this newline before end-of-file comments.
291   if (Style.Language == FormatStyle::LK_TextProto &&
292       !CommentsBeforeNextToken.empty()) {
293     addUnwrappedLine();
294   }
295   flushComments(true);
296   addUnwrappedLine();
297 }
298 
299 void UnwrappedLineParser::parseCSharpGenericTypeConstraint() {
300   do {
301     switch (FormatTok->Tok.getKind()) {
302     case tok::l_brace:
303       return;
304     default:
305       if (FormatTok->is(Keywords.kw_where)) {
306         addUnwrappedLine();
307         nextToken();
308         parseCSharpGenericTypeConstraint();
309         break;
310       }
311       nextToken();
312       break;
313     }
314   } while (!eof());
315 }
316 
317 void UnwrappedLineParser::parseCSharpAttribute() {
318   int UnpairedSquareBrackets = 1;
319   do {
320     switch (FormatTok->Tok.getKind()) {
321     case tok::r_square:
322       nextToken();
323       --UnpairedSquareBrackets;
324       if (UnpairedSquareBrackets == 0) {
325         addUnwrappedLine();
326         return;
327       }
328       break;
329     case tok::l_square:
330       ++UnpairedSquareBrackets;
331       nextToken();
332       break;
333     default:
334       nextToken();
335       break;
336     }
337   } while (!eof());
338 }
339 
340 bool UnwrappedLineParser::precededByCommentOrPPDirective() const {
341   if (!Lines.empty() && Lines.back().InPPDirective)
342     return true;
343 
344   const FormatToken *Previous = Tokens->getPreviousToken();
345   return Previous && Previous->is(tok::comment) &&
346          (Previous->IsMultiline || Previous->NewlinesBefore > 0);
347 }
348 
349 /// \brief Parses a level, that is ???.
350 /// \param OpeningBrace Opening brace (\p nullptr if absent) of that level.
351 /// \param IfKind The \p if statement kind in the level.
352 /// \param IfLeftBrace The left brace of the \p if block in the level.
353 /// \returns true if a simple block of if/else/for/while, or false otherwise.
354 /// (A simple block has a single statement.)
355 bool UnwrappedLineParser::parseLevel(const FormatToken *OpeningBrace,
356                                      IfStmtKind *IfKind,
357                                      FormatToken **IfLeftBrace) {
358   const bool InRequiresExpression =
359       OpeningBrace && OpeningBrace->is(TT_RequiresExpressionLBrace);
360   const bool IsPrecededByCommentOrPPDirective =
361       !Style.RemoveBracesLLVM || precededByCommentOrPPDirective();
362   FormatToken *IfLBrace = nullptr;
363   bool HasDoWhile = false;
364   bool HasLabel = false;
365   unsigned StatementCount = 0;
366   bool SwitchLabelEncountered = false;
367 
368   do {
369     if (FormatTok->isAttribute()) {
370       nextToken();
371       if (FormatTok->is(tok::l_paren))
372         parseParens();
373       continue;
374     }
375     tok::TokenKind Kind = FormatTok->Tok.getKind();
376     if (FormatTok->is(TT_MacroBlockBegin))
377       Kind = tok::l_brace;
378     else if (FormatTok->is(TT_MacroBlockEnd))
379       Kind = tok::r_brace;
380 
381     auto ParseDefault = [this, OpeningBrace, IfKind, &IfLBrace, &HasDoWhile,
382                          &HasLabel, &StatementCount] {
383       parseStructuralElement(OpeningBrace, IfKind, &IfLBrace,
384                              HasDoWhile ? nullptr : &HasDoWhile,
385                              HasLabel ? nullptr : &HasLabel);
386       ++StatementCount;
387       assert(StatementCount > 0 && "StatementCount overflow!");
388     };
389 
390     switch (Kind) {
391     case tok::comment:
392       nextToken();
393       addUnwrappedLine();
394       break;
395     case tok::l_brace:
396       if (InRequiresExpression) {
397         FormatTok->setFinalizedType(TT_RequiresExpressionLBrace);
398       } else if (FormatTok->Previous &&
399                  FormatTok->Previous->ClosesRequiresClause) {
400         // We need the 'default' case here to correctly parse a function
401         // l_brace.
402         ParseDefault();
403         continue;
404       }
405       if (!InRequiresExpression && FormatTok->isNot(TT_MacroBlockBegin)) {
406         if (tryToParseBracedList())
407           continue;
408         FormatTok->setFinalizedType(TT_BlockLBrace);
409       }
410       parseBlock();
411       ++StatementCount;
412       assert(StatementCount > 0 && "StatementCount overflow!");
413       addUnwrappedLine();
414       break;
415     case tok::r_brace:
416       if (OpeningBrace) {
417         if (!Style.RemoveBracesLLVM || Line->InPPDirective ||
418             !OpeningBrace->isOneOf(TT_ControlStatementLBrace, TT_ElseLBrace)) {
419           return false;
420         }
421         if (FormatTok->isNot(tok::r_brace) || StatementCount != 1 || HasLabel ||
422             HasDoWhile || IsPrecededByCommentOrPPDirective ||
423             precededByCommentOrPPDirective()) {
424           return false;
425         }
426         const FormatToken *Next = Tokens->peekNextToken();
427         if (Next->is(tok::comment) && Next->NewlinesBefore == 0)
428           return false;
429         if (IfLeftBrace)
430           *IfLeftBrace = IfLBrace;
431         return true;
432       }
433       nextToken();
434       addUnwrappedLine();
435       break;
436     case tok::kw_default: {
437       unsigned StoredPosition = Tokens->getPosition();
438       auto *Next = Tokens->getNextNonComment();
439       FormatTok = Tokens->setPosition(StoredPosition);
440       if (!Next->isOneOf(tok::colon, tok::arrow)) {
441         // default not followed by `:` or `->` is not a case label; treat it
442         // like an identifier.
443         parseStructuralElement();
444         break;
445       }
446       // Else, if it is 'default:', fall through to the case handling.
447       [[fallthrough]];
448     }
449     case tok::kw_case:
450       if (Style.Language == FormatStyle::LK_Proto || Style.isVerilog() ||
451           (Style.isJavaScript() && Line->MustBeDeclaration)) {
452         // Proto: there are no switch/case statements
453         // Verilog: Case labels don't have this word. We handle case
454         // labels including default in TokenAnnotator.
455         // JavaScript: A 'case: string' style field declaration.
456         ParseDefault();
457         break;
458       }
459       if (!SwitchLabelEncountered &&
460           (Style.IndentCaseLabels ||
461            (OpeningBrace && OpeningBrace->is(TT_SwitchExpressionLBrace)) ||
462            (Line->InPPDirective && Line->Level == 1))) {
463         ++Line->Level;
464       }
465       SwitchLabelEncountered = true;
466       parseStructuralElement();
467       break;
468     case tok::l_square:
469       if (Style.isCSharp()) {
470         nextToken();
471         parseCSharpAttribute();
472         break;
473       }
474       if (handleCppAttributes())
475         break;
476       [[fallthrough]];
477     default:
478       ParseDefault();
479       break;
480     }
481   } while (!eof());
482 
483   return false;
484 }
485 
486 void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) {
487   // We'll parse forward through the tokens until we hit
488   // a closing brace or eof - note that getNextToken() will
489   // parse macros, so this will magically work inside macro
490   // definitions, too.
491   unsigned StoredPosition = Tokens->getPosition();
492   FormatToken *Tok = FormatTok;
493   const FormatToken *PrevTok = Tok->Previous;
494   // Keep a stack of positions of lbrace tokens. We will
495   // update information about whether an lbrace starts a
496   // braced init list or a different block during the loop.
497   struct StackEntry {
498     FormatToken *Tok;
499     const FormatToken *PrevTok;
500   };
501   SmallVector<StackEntry, 8> LBraceStack;
502   assert(Tok->is(tok::l_brace));
503 
504   do {
505     auto *NextTok = Tokens->getNextNonComment();
506 
507     if (!Line->InMacroBody && !Style.isTableGen()) {
508       // Skip PPDirective lines and comments.
509       while (NextTok->is(tok::hash)) {
510         NextTok = Tokens->getNextToken();
511         if (NextTok->is(tok::pp_not_keyword))
512           break;
513         do {
514           NextTok = Tokens->getNextToken();
515         } while (NextTok->NewlinesBefore == 0 && NextTok->isNot(tok::eof));
516 
517         while (NextTok->is(tok::comment))
518           NextTok = Tokens->getNextToken();
519       }
520     }
521 
522     switch (Tok->Tok.getKind()) {
523     case tok::l_brace:
524       if (Style.isJavaScript() && PrevTok) {
525         if (PrevTok->isOneOf(tok::colon, tok::less)) {
526           // A ':' indicates this code is in a type, or a braced list
527           // following a label in an object literal ({a: {b: 1}}).
528           // A '<' could be an object used in a comparison, but that is nonsense
529           // code (can never return true), so more likely it is a generic type
530           // argument (`X<{a: string; b: number}>`).
531           // The code below could be confused by semicolons between the
532           // individual members in a type member list, which would normally
533           // trigger BK_Block. In both cases, this must be parsed as an inline
534           // braced init.
535           Tok->setBlockKind(BK_BracedInit);
536         } else if (PrevTok->is(tok::r_paren)) {
537           // `) { }` can only occur in function or method declarations in JS.
538           Tok->setBlockKind(BK_Block);
539         }
540       } else {
541         Tok->setBlockKind(BK_Unknown);
542       }
543       LBraceStack.push_back({Tok, PrevTok});
544       break;
545     case tok::r_brace:
546       if (LBraceStack.empty())
547         break;
548       if (auto *LBrace = LBraceStack.back().Tok; LBrace->is(BK_Unknown)) {
549         bool ProbablyBracedList = false;
550         if (Style.Language == FormatStyle::LK_Proto) {
551           ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square);
552         } else if (LBrace->isNot(TT_EnumLBrace)) {
553           // Using OriginalColumn to distinguish between ObjC methods and
554           // binary operators is a bit hacky.
555           bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) &&
556                                   NextTok->OriginalColumn == 0;
557 
558           // Try to detect a braced list. Note that regardless how we mark inner
559           // braces here, we will overwrite the BlockKind later if we parse a
560           // braced list (where all blocks inside are by default braced lists),
561           // or when we explicitly detect blocks (for example while parsing
562           // lambdas).
563 
564           // If we already marked the opening brace as braced list, the closing
565           // must also be part of it.
566           ProbablyBracedList = LBrace->is(TT_BracedListLBrace);
567 
568           ProbablyBracedList = ProbablyBracedList ||
569                                (Style.isJavaScript() &&
570                                 NextTok->isOneOf(Keywords.kw_of, Keywords.kw_in,
571                                                  Keywords.kw_as));
572           ProbablyBracedList =
573               ProbablyBracedList || (IsCpp && (PrevTok->Tok.isLiteral() ||
574                                                NextTok->is(tok::l_paren)));
575 
576           // If there is a comma, semicolon or right paren after the closing
577           // brace, we assume this is a braced initializer list.
578           // FIXME: Some of these do not apply to JS, e.g. "} {" can never be a
579           // braced list in JS.
580           ProbablyBracedList =
581               ProbablyBracedList ||
582               NextTok->isOneOf(tok::comma, tok::period, tok::colon,
583                                tok::r_paren, tok::r_square, tok::ellipsis);
584 
585           // Distinguish between braced list in a constructor initializer list
586           // followed by constructor body, or just adjacent blocks.
587           ProbablyBracedList =
588               ProbablyBracedList ||
589               (NextTok->is(tok::l_brace) && LBraceStack.back().PrevTok &&
590                LBraceStack.back().PrevTok->isOneOf(tok::identifier,
591                                                    tok::greater));
592 
593           ProbablyBracedList =
594               ProbablyBracedList ||
595               (NextTok->is(tok::identifier) &&
596                !PrevTok->isOneOf(tok::semi, tok::r_brace, tok::l_brace));
597 
598           ProbablyBracedList = ProbablyBracedList ||
599                                (NextTok->is(tok::semi) &&
600                                 (!ExpectClassBody || LBraceStack.size() != 1));
601 
602           ProbablyBracedList =
603               ProbablyBracedList ||
604               (NextTok->isBinaryOperator() && !NextIsObjCMethod);
605 
606           if (!Style.isCSharp() && NextTok->is(tok::l_square)) {
607             // We can have an array subscript after a braced init
608             // list, but C++11 attributes are expected after blocks.
609             NextTok = Tokens->getNextToken();
610             ProbablyBracedList = NextTok->isNot(tok::l_square);
611           }
612 
613           // Cpp macro definition body that is a nonempty braced list or block:
614           if (IsCpp && Line->InMacroBody && PrevTok != FormatTok &&
615               !FormatTok->Previous && NextTok->is(tok::eof) &&
616               // A statement can end with only `;` (simple statement), a block
617               // closing brace (compound statement), or `:` (label statement).
618               // If PrevTok is a block opening brace, Tok ends an empty block.
619               !PrevTok->isOneOf(tok::semi, BK_Block, tok::colon)) {
620             ProbablyBracedList = true;
621           }
622         }
623         const auto BlockKind = ProbablyBracedList ? BK_BracedInit : BK_Block;
624         Tok->setBlockKind(BlockKind);
625         LBrace->setBlockKind(BlockKind);
626       }
627       LBraceStack.pop_back();
628       break;
629     case tok::identifier:
630       if (Tok->isNot(TT_StatementMacro))
631         break;
632       [[fallthrough]];
633     case tok::at:
634     case tok::semi:
635     case tok::kw_if:
636     case tok::kw_while:
637     case tok::kw_for:
638     case tok::kw_switch:
639     case tok::kw_try:
640     case tok::kw___try:
641       if (!LBraceStack.empty() && LBraceStack.back().Tok->is(BK_Unknown))
642         LBraceStack.back().Tok->setBlockKind(BK_Block);
643       break;
644     default:
645       break;
646     }
647 
648     PrevTok = Tok;
649     Tok = NextTok;
650   } while (Tok->isNot(tok::eof) && !LBraceStack.empty());
651 
652   // Assume other blocks for all unclosed opening braces.
653   for (const auto &Entry : LBraceStack)
654     if (Entry.Tok->is(BK_Unknown))
655       Entry.Tok->setBlockKind(BK_Block);
656 
657   FormatTok = Tokens->setPosition(StoredPosition);
658 }
659 
660 // Sets the token type of the directly previous right brace.
661 void UnwrappedLineParser::setPreviousRBraceType(TokenType Type) {
662   if (auto Prev = FormatTok->getPreviousNonComment();
663       Prev && Prev->is(tok::r_brace)) {
664     Prev->setFinalizedType(Type);
665   }
666 }
667 
668 template <class T>
669 static inline void hash_combine(std::size_t &seed, const T &v) {
670   std::hash<T> hasher;
671   seed ^= hasher(v) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
672 }
673 
674 size_t UnwrappedLineParser::computePPHash() const {
675   size_t h = 0;
676   for (const auto &i : PPStack) {
677     hash_combine(h, size_t(i.Kind));
678     hash_combine(h, i.Line);
679   }
680   return h;
681 }
682 
683 // Checks whether \p ParsedLine might fit on a single line. If \p OpeningBrace
684 // is not null, subtracts its length (plus the preceding space) when computing
685 // the length of \p ParsedLine. We must clone the tokens of \p ParsedLine before
686 // running the token annotator on it so that we can restore them afterward.
687 bool UnwrappedLineParser::mightFitOnOneLine(
688     UnwrappedLine &ParsedLine, const FormatToken *OpeningBrace) const {
689   const auto ColumnLimit = Style.ColumnLimit;
690   if (ColumnLimit == 0)
691     return true;
692 
693   auto &Tokens = ParsedLine.Tokens;
694   assert(!Tokens.empty());
695 
696   const auto *LastToken = Tokens.back().Tok;
697   assert(LastToken);
698 
699   SmallVector<UnwrappedLineNode> SavedTokens(Tokens.size());
700 
701   int Index = 0;
702   for (const auto &Token : Tokens) {
703     assert(Token.Tok);
704     auto &SavedToken = SavedTokens[Index++];
705     SavedToken.Tok = new FormatToken;
706     SavedToken.Tok->copyFrom(*Token.Tok);
707     SavedToken.Children = std::move(Token.Children);
708   }
709 
710   AnnotatedLine Line(ParsedLine);
711   assert(Line.Last == LastToken);
712 
713   TokenAnnotator Annotator(Style, Keywords);
714   Annotator.annotate(Line);
715   Annotator.calculateFormattingInformation(Line);
716 
717   auto Length = LastToken->TotalLength;
718   if (OpeningBrace) {
719     assert(OpeningBrace != Tokens.front().Tok);
720     if (auto Prev = OpeningBrace->Previous;
721         Prev && Prev->TotalLength + ColumnLimit == OpeningBrace->TotalLength) {
722       Length -= ColumnLimit;
723     }
724     Length -= OpeningBrace->TokenText.size() + 1;
725   }
726 
727   if (const auto *FirstToken = Line.First; FirstToken->is(tok::r_brace)) {
728     assert(!OpeningBrace || OpeningBrace->is(TT_ControlStatementLBrace));
729     Length -= FirstToken->TokenText.size() + 1;
730   }
731 
732   Index = 0;
733   for (auto &Token : Tokens) {
734     const auto &SavedToken = SavedTokens[Index++];
735     Token.Tok->copyFrom(*SavedToken.Tok);
736     Token.Children = std::move(SavedToken.Children);
737     delete SavedToken.Tok;
738   }
739 
740   // If these change PPLevel needs to be used for get correct indentation.
741   assert(!Line.InMacroBody);
742   assert(!Line.InPPDirective);
743   return Line.Level * Style.IndentWidth + Length <= ColumnLimit;
744 }
745 
746 FormatToken *UnwrappedLineParser::parseBlock(bool MustBeDeclaration,
747                                              unsigned AddLevels, bool MunchSemi,
748                                              bool KeepBraces,
749                                              IfStmtKind *IfKind,
750                                              bool UnindentWhitesmithsBraces) {
751   auto HandleVerilogBlockLabel = [this]() {
752     // ":" name
753     if (Style.isVerilog() && FormatTok->is(tok::colon)) {
754       nextToken();
755       if (Keywords.isVerilogIdentifier(*FormatTok))
756         nextToken();
757     }
758   };
759 
760   // Whether this is a Verilog-specific block that has a special header like a
761   // module.
762   const bool VerilogHierarchy =
763       Style.isVerilog() && Keywords.isVerilogHierarchy(*FormatTok);
764   assert((FormatTok->isOneOf(tok::l_brace, TT_MacroBlockBegin) ||
765           (Style.isVerilog() &&
766            (Keywords.isVerilogBegin(*FormatTok) || VerilogHierarchy))) &&
767          "'{' or macro block token expected");
768   FormatToken *Tok = FormatTok;
769   const bool FollowedByComment = Tokens->peekNextToken()->is(tok::comment);
770   auto Index = CurrentLines->size();
771   const bool MacroBlock = FormatTok->is(TT_MacroBlockBegin);
772   FormatTok->setBlockKind(BK_Block);
773 
774   // For Whitesmiths mode, jump to the next level prior to skipping over the
775   // braces.
776   if (!VerilogHierarchy && AddLevels > 0 &&
777       Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) {
778     ++Line->Level;
779   }
780 
781   size_t PPStartHash = computePPHash();
782 
783   const unsigned InitialLevel = Line->Level;
784   if (VerilogHierarchy) {
785     AddLevels += parseVerilogHierarchyHeader();
786   } else {
787     nextToken(/*LevelDifference=*/AddLevels);
788     HandleVerilogBlockLabel();
789   }
790 
791   // Bail out if there are too many levels. Otherwise, the stack might overflow.
792   if (Line->Level > 300)
793     return nullptr;
794 
795   if (MacroBlock && FormatTok->is(tok::l_paren))
796     parseParens();
797 
798   size_t NbPreprocessorDirectives =
799       !parsingPPDirective() ? PreprocessorDirectives.size() : 0;
800   addUnwrappedLine();
801   size_t OpeningLineIndex =
802       CurrentLines->empty()
803           ? (UnwrappedLine::kInvalidIndex)
804           : (CurrentLines->size() - 1 - NbPreprocessorDirectives);
805 
806   // Whitesmiths is weird here. The brace needs to be indented for the namespace
807   // block, but the block itself may not be indented depending on the style
808   // settings. This allows the format to back up one level in those cases.
809   if (UnindentWhitesmithsBraces)
810     --Line->Level;
811 
812   ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
813                                           MustBeDeclaration);
814   if (AddLevels > 0u && Style.BreakBeforeBraces != FormatStyle::BS_Whitesmiths)
815     Line->Level += AddLevels;
816 
817   FormatToken *IfLBrace = nullptr;
818   const bool SimpleBlock = parseLevel(Tok, IfKind, &IfLBrace);
819 
820   if (eof())
821     return IfLBrace;
822 
823   if (MacroBlock ? FormatTok->isNot(TT_MacroBlockEnd)
824                  : FormatTok->isNot(tok::r_brace)) {
825     Line->Level = InitialLevel;
826     FormatTok->setBlockKind(BK_Block);
827     return IfLBrace;
828   }
829 
830   if (FormatTok->is(tok::r_brace)) {
831     FormatTok->setBlockKind(BK_Block);
832     if (Tok->is(TT_NamespaceLBrace))
833       FormatTok->setFinalizedType(TT_NamespaceRBrace);
834   }
835 
836   const bool IsFunctionRBrace =
837       FormatTok->is(tok::r_brace) && Tok->is(TT_FunctionLBrace);
838 
839   auto RemoveBraces = [=]() mutable {
840     if (!SimpleBlock)
841       return false;
842     assert(Tok->isOneOf(TT_ControlStatementLBrace, TT_ElseLBrace));
843     assert(FormatTok->is(tok::r_brace));
844     const bool WrappedOpeningBrace = !Tok->Previous;
845     if (WrappedOpeningBrace && FollowedByComment)
846       return false;
847     const bool HasRequiredIfBraces = IfLBrace && !IfLBrace->Optional;
848     if (KeepBraces && !HasRequiredIfBraces)
849       return false;
850     if (Tok->isNot(TT_ElseLBrace) || !HasRequiredIfBraces) {
851       const FormatToken *Previous = Tokens->getPreviousToken();
852       assert(Previous);
853       if (Previous->is(tok::r_brace) && !Previous->Optional)
854         return false;
855     }
856     assert(!CurrentLines->empty());
857     auto &LastLine = CurrentLines->back();
858     if (LastLine.Level == InitialLevel + 1 && !mightFitOnOneLine(LastLine))
859       return false;
860     if (Tok->is(TT_ElseLBrace))
861       return true;
862     if (WrappedOpeningBrace) {
863       assert(Index > 0);
864       --Index; // The line above the wrapped l_brace.
865       Tok = nullptr;
866     }
867     return mightFitOnOneLine((*CurrentLines)[Index], Tok);
868   };
869   if (RemoveBraces()) {
870     Tok->MatchingParen = FormatTok;
871     FormatTok->MatchingParen = Tok;
872   }
873 
874   size_t PPEndHash = computePPHash();
875 
876   // Munch the closing brace.
877   nextToken(/*LevelDifference=*/-AddLevels);
878 
879   // When this is a function block and there is an unnecessary semicolon
880   // afterwards then mark it as optional (so the RemoveSemi pass can get rid of
881   // it later).
882   if (Style.RemoveSemicolon && IsFunctionRBrace) {
883     while (FormatTok->is(tok::semi)) {
884       FormatTok->Optional = true;
885       nextToken();
886     }
887   }
888 
889   HandleVerilogBlockLabel();
890 
891   if (MacroBlock && FormatTok->is(tok::l_paren))
892     parseParens();
893 
894   Line->Level = InitialLevel;
895 
896   if (FormatTok->is(tok::kw_noexcept)) {
897     // A noexcept in a requires expression.
898     nextToken();
899   }
900 
901   if (FormatTok->is(tok::arrow)) {
902     // Following the } or noexcept we can find a trailing return type arrow
903     // as part of an implicit conversion constraint.
904     nextToken();
905     parseStructuralElement();
906   }
907 
908   if (MunchSemi && FormatTok->is(tok::semi))
909     nextToken();
910 
911   if (PPStartHash == PPEndHash) {
912     Line->MatchingOpeningBlockLineIndex = OpeningLineIndex;
913     if (OpeningLineIndex != UnwrappedLine::kInvalidIndex) {
914       // Update the opening line to add the forward reference as well
915       (*CurrentLines)[OpeningLineIndex].MatchingClosingBlockLineIndex =
916           CurrentLines->size() - 1;
917     }
918   }
919 
920   return IfLBrace;
921 }
922 
923 static bool isGoogScope(const UnwrappedLine &Line) {
924   // FIXME: Closure-library specific stuff should not be hard-coded but be
925   // configurable.
926   if (Line.Tokens.size() < 4)
927     return false;
928   auto I = Line.Tokens.begin();
929   if (I->Tok->TokenText != "goog")
930     return false;
931   ++I;
932   if (I->Tok->isNot(tok::period))
933     return false;
934   ++I;
935   if (I->Tok->TokenText != "scope")
936     return false;
937   ++I;
938   return I->Tok->is(tok::l_paren);
939 }
940 
941 static bool isIIFE(const UnwrappedLine &Line,
942                    const AdditionalKeywords &Keywords) {
943   // Look for the start of an immediately invoked anonymous function.
944   // https://en.wikipedia.org/wiki/Immediately-invoked_function_expression
945   // This is commonly done in JavaScript to create a new, anonymous scope.
946   // Example: (function() { ... })()
947   if (Line.Tokens.size() < 3)
948     return false;
949   auto I = Line.Tokens.begin();
950   if (I->Tok->isNot(tok::l_paren))
951     return false;
952   ++I;
953   if (I->Tok->isNot(Keywords.kw_function))
954     return false;
955   ++I;
956   return I->Tok->is(tok::l_paren);
957 }
958 
959 static bool ShouldBreakBeforeBrace(const FormatStyle &Style,
960                                    const FormatToken &InitialToken) {
961   tok::TokenKind Kind = InitialToken.Tok.getKind();
962   if (InitialToken.is(TT_NamespaceMacro))
963     Kind = tok::kw_namespace;
964 
965   switch (Kind) {
966   case tok::kw_namespace:
967     return Style.BraceWrapping.AfterNamespace;
968   case tok::kw_class:
969     return Style.BraceWrapping.AfterClass;
970   case tok::kw_union:
971     return Style.BraceWrapping.AfterUnion;
972   case tok::kw_struct:
973     return Style.BraceWrapping.AfterStruct;
974   case tok::kw_enum:
975     return Style.BraceWrapping.AfterEnum;
976   default:
977     return false;
978   }
979 }
980 
981 void UnwrappedLineParser::parseChildBlock() {
982   assert(FormatTok->is(tok::l_brace));
983   FormatTok->setBlockKind(BK_Block);
984   const FormatToken *OpeningBrace = FormatTok;
985   nextToken();
986   {
987     bool SkipIndent = (Style.isJavaScript() &&
988                        (isGoogScope(*Line) || isIIFE(*Line, Keywords)));
989     ScopedLineState LineState(*this);
990     ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
991                                             /*MustBeDeclaration=*/false);
992     Line->Level += SkipIndent ? 0 : 1;
993     parseLevel(OpeningBrace);
994     flushComments(isOnNewLine(*FormatTok));
995     Line->Level -= SkipIndent ? 0 : 1;
996   }
997   nextToken();
998 }
999 
1000 void UnwrappedLineParser::parsePPDirective() {
1001   assert(FormatTok->is(tok::hash) && "'#' expected");
1002   ScopedMacroState MacroState(*Line, Tokens, FormatTok);
1003 
1004   nextToken();
1005 
1006   if (!FormatTok->Tok.getIdentifierInfo()) {
1007     parsePPUnknown();
1008     return;
1009   }
1010 
1011   switch (FormatTok->Tok.getIdentifierInfo()->getPPKeywordID()) {
1012   case tok::pp_define:
1013     parsePPDefine();
1014     return;
1015   case tok::pp_if:
1016     parsePPIf(/*IfDef=*/false);
1017     break;
1018   case tok::pp_ifdef:
1019   case tok::pp_ifndef:
1020     parsePPIf(/*IfDef=*/true);
1021     break;
1022   case tok::pp_else:
1023   case tok::pp_elifdef:
1024   case tok::pp_elifndef:
1025   case tok::pp_elif:
1026     parsePPElse();
1027     break;
1028   case tok::pp_endif:
1029     parsePPEndIf();
1030     break;
1031   case tok::pp_pragma:
1032     parsePPPragma();
1033     break;
1034   default:
1035     parsePPUnknown();
1036     break;
1037   }
1038 }
1039 
1040 void UnwrappedLineParser::conditionalCompilationCondition(bool Unreachable) {
1041   size_t Line = CurrentLines->size();
1042   if (CurrentLines == &PreprocessorDirectives)
1043     Line += Lines.size();
1044 
1045   if (Unreachable ||
1046       (!PPStack.empty() && PPStack.back().Kind == PP_Unreachable)) {
1047     PPStack.push_back({PP_Unreachable, Line});
1048   } else {
1049     PPStack.push_back({PP_Conditional, Line});
1050   }
1051 }
1052 
1053 void UnwrappedLineParser::conditionalCompilationStart(bool Unreachable) {
1054   ++PPBranchLevel;
1055   assert(PPBranchLevel >= 0 && PPBranchLevel <= (int)PPLevelBranchIndex.size());
1056   if (PPBranchLevel == (int)PPLevelBranchIndex.size()) {
1057     PPLevelBranchIndex.push_back(0);
1058     PPLevelBranchCount.push_back(0);
1059   }
1060   PPChainBranchIndex.push(Unreachable ? -1 : 0);
1061   bool Skip = PPLevelBranchIndex[PPBranchLevel] > 0;
1062   conditionalCompilationCondition(Unreachable || Skip);
1063 }
1064 
1065 void UnwrappedLineParser::conditionalCompilationAlternative() {
1066   if (!PPStack.empty())
1067     PPStack.pop_back();
1068   assert(PPBranchLevel < (int)PPLevelBranchIndex.size());
1069   if (!PPChainBranchIndex.empty())
1070     ++PPChainBranchIndex.top();
1071   conditionalCompilationCondition(
1072       PPBranchLevel >= 0 && !PPChainBranchIndex.empty() &&
1073       PPLevelBranchIndex[PPBranchLevel] != PPChainBranchIndex.top());
1074 }
1075 
1076 void UnwrappedLineParser::conditionalCompilationEnd() {
1077   assert(PPBranchLevel < (int)PPLevelBranchIndex.size());
1078   if (PPBranchLevel >= 0 && !PPChainBranchIndex.empty()) {
1079     if (PPChainBranchIndex.top() + 1 > PPLevelBranchCount[PPBranchLevel])
1080       PPLevelBranchCount[PPBranchLevel] = PPChainBranchIndex.top() + 1;
1081   }
1082   // Guard against #endif's without #if.
1083   if (PPBranchLevel > -1)
1084     --PPBranchLevel;
1085   if (!PPChainBranchIndex.empty())
1086     PPChainBranchIndex.pop();
1087   if (!PPStack.empty())
1088     PPStack.pop_back();
1089 }
1090 
1091 void UnwrappedLineParser::parsePPIf(bool IfDef) {
1092   bool IfNDef = FormatTok->is(tok::pp_ifndef);
1093   nextToken();
1094   bool Unreachable = false;
1095   if (!IfDef && (FormatTok->is(tok::kw_false) || FormatTok->TokenText == "0"))
1096     Unreachable = true;
1097   if (IfDef && !IfNDef && FormatTok->TokenText == "SWIG")
1098     Unreachable = true;
1099   conditionalCompilationStart(Unreachable);
1100   FormatToken *IfCondition = FormatTok;
1101   // If there's a #ifndef on the first line, and the only lines before it are
1102   // comments, it could be an include guard.
1103   bool MaybeIncludeGuard = IfNDef;
1104   if (IncludeGuard == IG_Inited && MaybeIncludeGuard) {
1105     for (auto &Line : Lines) {
1106       if (Line.Tokens.front().Tok->isNot(tok::comment)) {
1107         MaybeIncludeGuard = false;
1108         IncludeGuard = IG_Rejected;
1109         break;
1110       }
1111     }
1112   }
1113   --PPBranchLevel;
1114   parsePPUnknown();
1115   ++PPBranchLevel;
1116   if (IncludeGuard == IG_Inited && MaybeIncludeGuard) {
1117     IncludeGuard = IG_IfNdefed;
1118     IncludeGuardToken = IfCondition;
1119   }
1120 }
1121 
1122 void UnwrappedLineParser::parsePPElse() {
1123   // If a potential include guard has an #else, it's not an include guard.
1124   if (IncludeGuard == IG_Defined && PPBranchLevel == 0)
1125     IncludeGuard = IG_Rejected;
1126   // Don't crash when there is an #else without an #if.
1127   assert(PPBranchLevel >= -1);
1128   if (PPBranchLevel == -1)
1129     conditionalCompilationStart(/*Unreachable=*/true);
1130   conditionalCompilationAlternative();
1131   --PPBranchLevel;
1132   parsePPUnknown();
1133   ++PPBranchLevel;
1134 }
1135 
1136 void UnwrappedLineParser::parsePPEndIf() {
1137   conditionalCompilationEnd();
1138   parsePPUnknown();
1139   // If the #endif of a potential include guard is the last thing in the file,
1140   // then we found an include guard.
1141   if (IncludeGuard == IG_Defined && PPBranchLevel == -1 && Tokens->isEOF() &&
1142       Style.IndentPPDirectives != FormatStyle::PPDIS_None) {
1143     IncludeGuard = IG_Found;
1144   }
1145 }
1146 
1147 void UnwrappedLineParser::parsePPDefine() {
1148   nextToken();
1149 
1150   if (!FormatTok->Tok.getIdentifierInfo()) {
1151     IncludeGuard = IG_Rejected;
1152     IncludeGuardToken = nullptr;
1153     parsePPUnknown();
1154     return;
1155   }
1156 
1157   if (IncludeGuard == IG_IfNdefed &&
1158       IncludeGuardToken->TokenText == FormatTok->TokenText) {
1159     IncludeGuard = IG_Defined;
1160     IncludeGuardToken = nullptr;
1161     for (auto &Line : Lines) {
1162       if (!Line.Tokens.front().Tok->isOneOf(tok::comment, tok::hash)) {
1163         IncludeGuard = IG_Rejected;
1164         break;
1165       }
1166     }
1167   }
1168 
1169   // In the context of a define, even keywords should be treated as normal
1170   // identifiers. Setting the kind to identifier is not enough, because we need
1171   // to treat additional keywords like __except as well, which are already
1172   // identifiers. Setting the identifier info to null interferes with include
1173   // guard processing above, and changes preprocessing nesting.
1174   FormatTok->Tok.setKind(tok::identifier);
1175   FormatTok->Tok.setIdentifierInfo(Keywords.kw_internal_ident_after_define);
1176   nextToken();
1177   if (FormatTok->Tok.getKind() == tok::l_paren &&
1178       !FormatTok->hasWhitespaceBefore()) {
1179     parseParens();
1180   }
1181   if (Style.IndentPPDirectives != FormatStyle::PPDIS_None)
1182     Line->Level += PPBranchLevel + 1;
1183   addUnwrappedLine();
1184   ++Line->Level;
1185 
1186   Line->PPLevel = PPBranchLevel + (IncludeGuard == IG_Defined ? 0 : 1);
1187   assert((int)Line->PPLevel >= 0);
1188   Line->InMacroBody = true;
1189 
1190   if (Style.SkipMacroDefinitionBody) {
1191     while (!eof()) {
1192       FormatTok->Finalized = true;
1193       FormatTok = Tokens->getNextToken();
1194     }
1195     addUnwrappedLine();
1196     return;
1197   }
1198 
1199   // Errors during a preprocessor directive can only affect the layout of the
1200   // preprocessor directive, and thus we ignore them. An alternative approach
1201   // would be to use the same approach we use on the file level (no
1202   // re-indentation if there was a structural error) within the macro
1203   // definition.
1204   parseFile();
1205 }
1206 
1207 void UnwrappedLineParser::parsePPPragma() {
1208   Line->InPragmaDirective = true;
1209   parsePPUnknown();
1210 }
1211 
1212 void UnwrappedLineParser::parsePPUnknown() {
1213   do {
1214     nextToken();
1215   } while (!eof());
1216   if (Style.IndentPPDirectives != FormatStyle::PPDIS_None)
1217     Line->Level += PPBranchLevel + 1;
1218   addUnwrappedLine();
1219 }
1220 
1221 // Here we exclude certain tokens that are not usually the first token in an
1222 // unwrapped line. This is used in attempt to distinguish macro calls without
1223 // trailing semicolons from other constructs split to several lines.
1224 static bool tokenCanStartNewLine(const FormatToken &Tok) {
1225   // Semicolon can be a null-statement, l_square can be a start of a macro or
1226   // a C++11 attribute, but this doesn't seem to be common.
1227   return !Tok.isOneOf(tok::semi, tok::l_brace,
1228                       // Tokens that can only be used as binary operators and a
1229                       // part of overloaded operator names.
1230                       tok::period, tok::periodstar, tok::arrow, tok::arrowstar,
1231                       tok::less, tok::greater, tok::slash, tok::percent,
1232                       tok::lessless, tok::greatergreater, tok::equal,
1233                       tok::plusequal, tok::minusequal, tok::starequal,
1234                       tok::slashequal, tok::percentequal, tok::ampequal,
1235                       tok::pipeequal, tok::caretequal, tok::greatergreaterequal,
1236                       tok::lesslessequal,
1237                       // Colon is used in labels, base class lists, initializer
1238                       // lists, range-based for loops, ternary operator, but
1239                       // should never be the first token in an unwrapped line.
1240                       tok::colon,
1241                       // 'noexcept' is a trailing annotation.
1242                       tok::kw_noexcept);
1243 }
1244 
1245 static bool mustBeJSIdent(const AdditionalKeywords &Keywords,
1246                           const FormatToken *FormatTok) {
1247   // FIXME: This returns true for C/C++ keywords like 'struct'.
1248   return FormatTok->is(tok::identifier) &&
1249          (!FormatTok->Tok.getIdentifierInfo() ||
1250           !FormatTok->isOneOf(
1251               Keywords.kw_in, Keywords.kw_of, Keywords.kw_as, Keywords.kw_async,
1252               Keywords.kw_await, Keywords.kw_yield, Keywords.kw_finally,
1253               Keywords.kw_function, Keywords.kw_import, Keywords.kw_is,
1254               Keywords.kw_let, Keywords.kw_var, tok::kw_const,
1255               Keywords.kw_abstract, Keywords.kw_extends, Keywords.kw_implements,
1256               Keywords.kw_instanceof, Keywords.kw_interface,
1257               Keywords.kw_override, Keywords.kw_throws, Keywords.kw_from));
1258 }
1259 
1260 static bool mustBeJSIdentOrValue(const AdditionalKeywords &Keywords,
1261                                  const FormatToken *FormatTok) {
1262   return FormatTok->Tok.isLiteral() ||
1263          FormatTok->isOneOf(tok::kw_true, tok::kw_false) ||
1264          mustBeJSIdent(Keywords, FormatTok);
1265 }
1266 
1267 // isJSDeclOrStmt returns true if |FormatTok| starts a declaration or statement
1268 // when encountered after a value (see mustBeJSIdentOrValue).
1269 static bool isJSDeclOrStmt(const AdditionalKeywords &Keywords,
1270                            const FormatToken *FormatTok) {
1271   return FormatTok->isOneOf(
1272       tok::kw_return, Keywords.kw_yield,
1273       // conditionals
1274       tok::kw_if, tok::kw_else,
1275       // loops
1276       tok::kw_for, tok::kw_while, tok::kw_do, tok::kw_continue, tok::kw_break,
1277       // switch/case
1278       tok::kw_switch, tok::kw_case,
1279       // exceptions
1280       tok::kw_throw, tok::kw_try, tok::kw_catch, Keywords.kw_finally,
1281       // declaration
1282       tok::kw_const, tok::kw_class, Keywords.kw_var, Keywords.kw_let,
1283       Keywords.kw_async, Keywords.kw_function,
1284       // import/export
1285       Keywords.kw_import, tok::kw_export);
1286 }
1287 
1288 // Checks whether a token is a type in K&R C (aka C78).
1289 static bool isC78Type(const FormatToken &Tok) {
1290   return Tok.isOneOf(tok::kw_char, tok::kw_short, tok::kw_int, tok::kw_long,
1291                      tok::kw_unsigned, tok::kw_float, tok::kw_double,
1292                      tok::identifier);
1293 }
1294 
1295 // This function checks whether a token starts the first parameter declaration
1296 // in a K&R C (aka C78) function definition, e.g.:
1297 //   int f(a, b)
1298 //   short a, b;
1299 //   {
1300 //      return a + b;
1301 //   }
1302 static bool isC78ParameterDecl(const FormatToken *Tok, const FormatToken *Next,
1303                                const FormatToken *FuncName) {
1304   assert(Tok);
1305   assert(Next);
1306   assert(FuncName);
1307 
1308   if (FuncName->isNot(tok::identifier))
1309     return false;
1310 
1311   const FormatToken *Prev = FuncName->Previous;
1312   if (!Prev || (Prev->isNot(tok::star) && !isC78Type(*Prev)))
1313     return false;
1314 
1315   if (!isC78Type(*Tok) &&
1316       !Tok->isOneOf(tok::kw_register, tok::kw_struct, tok::kw_union)) {
1317     return false;
1318   }
1319 
1320   if (Next->isNot(tok::star) && !Next->Tok.getIdentifierInfo())
1321     return false;
1322 
1323   Tok = Tok->Previous;
1324   if (!Tok || Tok->isNot(tok::r_paren))
1325     return false;
1326 
1327   Tok = Tok->Previous;
1328   if (!Tok || Tok->isNot(tok::identifier))
1329     return false;
1330 
1331   return Tok->Previous && Tok->Previous->isOneOf(tok::l_paren, tok::comma);
1332 }
1333 
1334 bool UnwrappedLineParser::parseModuleImport() {
1335   assert(FormatTok->is(Keywords.kw_import) && "'import' expected");
1336 
1337   if (auto Token = Tokens->peekNextToken(/*SkipComment=*/true);
1338       !Token->Tok.getIdentifierInfo() &&
1339       !Token->isOneOf(tok::colon, tok::less, tok::string_literal)) {
1340     return false;
1341   }
1342 
1343   nextToken();
1344   while (!eof()) {
1345     if (FormatTok->is(tok::colon)) {
1346       FormatTok->setFinalizedType(TT_ModulePartitionColon);
1347     }
1348     // Handle import <foo/bar.h> as we would an include statement.
1349     else if (FormatTok->is(tok::less)) {
1350       nextToken();
1351       while (!FormatTok->isOneOf(tok::semi, tok::greater, tok::eof)) {
1352         // Mark tokens up to the trailing line comments as implicit string
1353         // literals.
1354         if (FormatTok->isNot(tok::comment) &&
1355             !FormatTok->TokenText.starts_with("//")) {
1356           FormatTok->setFinalizedType(TT_ImplicitStringLiteral);
1357         }
1358         nextToken();
1359       }
1360     }
1361     if (FormatTok->is(tok::semi)) {
1362       nextToken();
1363       break;
1364     }
1365     nextToken();
1366   }
1367 
1368   addUnwrappedLine();
1369   return true;
1370 }
1371 
1372 // readTokenWithJavaScriptASI reads the next token and terminates the current
1373 // line if JavaScript Automatic Semicolon Insertion must
1374 // happen between the current token and the next token.
1375 //
1376 // This method is conservative - it cannot cover all edge cases of JavaScript,
1377 // but only aims to correctly handle certain well known cases. It *must not*
1378 // return true in speculative cases.
1379 void UnwrappedLineParser::readTokenWithJavaScriptASI() {
1380   FormatToken *Previous = FormatTok;
1381   readToken();
1382   FormatToken *Next = FormatTok;
1383 
1384   bool IsOnSameLine =
1385       CommentsBeforeNextToken.empty()
1386           ? Next->NewlinesBefore == 0
1387           : CommentsBeforeNextToken.front()->NewlinesBefore == 0;
1388   if (IsOnSameLine)
1389     return;
1390 
1391   bool PreviousMustBeValue = mustBeJSIdentOrValue(Keywords, Previous);
1392   bool PreviousStartsTemplateExpr =
1393       Previous->is(TT_TemplateString) && Previous->TokenText.ends_with("${");
1394   if (PreviousMustBeValue || Previous->is(tok::r_paren)) {
1395     // If the line contains an '@' sign, the previous token might be an
1396     // annotation, which can precede another identifier/value.
1397     bool HasAt = llvm::any_of(Line->Tokens, [](UnwrappedLineNode &LineNode) {
1398       return LineNode.Tok->is(tok::at);
1399     });
1400     if (HasAt)
1401       return;
1402   }
1403   if (Next->is(tok::exclaim) && PreviousMustBeValue)
1404     return addUnwrappedLine();
1405   bool NextMustBeValue = mustBeJSIdentOrValue(Keywords, Next);
1406   bool NextEndsTemplateExpr =
1407       Next->is(TT_TemplateString) && Next->TokenText.starts_with("}");
1408   if (NextMustBeValue && !NextEndsTemplateExpr && !PreviousStartsTemplateExpr &&
1409       (PreviousMustBeValue ||
1410        Previous->isOneOf(tok::r_square, tok::r_paren, tok::plusplus,
1411                          tok::minusminus))) {
1412     return addUnwrappedLine();
1413   }
1414   if ((PreviousMustBeValue || Previous->is(tok::r_paren)) &&
1415       isJSDeclOrStmt(Keywords, Next)) {
1416     return addUnwrappedLine();
1417   }
1418 }
1419 
1420 void UnwrappedLineParser::parseStructuralElement(
1421     const FormatToken *OpeningBrace, IfStmtKind *IfKind,
1422     FormatToken **IfLeftBrace, bool *HasDoWhile, bool *HasLabel) {
1423   if (Style.Language == FormatStyle::LK_TableGen &&
1424       FormatTok->is(tok::pp_include)) {
1425     nextToken();
1426     if (FormatTok->is(tok::string_literal))
1427       nextToken();
1428     addUnwrappedLine();
1429     return;
1430   }
1431 
1432   if (IsCpp) {
1433     while (FormatTok->is(tok::l_square) && handleCppAttributes()) {
1434     }
1435   } else if (Style.isVerilog()) {
1436     if (Keywords.isVerilogStructuredProcedure(*FormatTok)) {
1437       parseForOrWhileLoop(/*HasParens=*/false);
1438       return;
1439     }
1440     if (FormatTok->isOneOf(Keywords.kw_foreach, Keywords.kw_repeat)) {
1441       parseForOrWhileLoop();
1442       return;
1443     }
1444     if (FormatTok->isOneOf(tok::kw_restrict, Keywords.kw_assert,
1445                            Keywords.kw_assume, Keywords.kw_cover)) {
1446       parseIfThenElse(IfKind, /*KeepBraces=*/false, /*IsVerilogAssert=*/true);
1447       return;
1448     }
1449 
1450     // Skip things that can exist before keywords like 'if' and 'case'.
1451     while (true) {
1452       if (FormatTok->isOneOf(Keywords.kw_priority, Keywords.kw_unique,
1453                              Keywords.kw_unique0)) {
1454         nextToken();
1455       } else if (FormatTok->is(tok::l_paren) &&
1456                  Tokens->peekNextToken()->is(tok::star)) {
1457         parseParens();
1458       } else {
1459         break;
1460       }
1461     }
1462   }
1463 
1464   // Tokens that only make sense at the beginning of a line.
1465   if (FormatTok->isAccessSpecifierKeyword()) {
1466     if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() ||
1467         Style.isCSharp()) {
1468       nextToken();
1469     } else {
1470       parseAccessSpecifier();
1471     }
1472     return;
1473   }
1474   switch (FormatTok->Tok.getKind()) {
1475   case tok::kw_asm:
1476     nextToken();
1477     if (FormatTok->is(tok::l_brace)) {
1478       FormatTok->setFinalizedType(TT_InlineASMBrace);
1479       nextToken();
1480       while (FormatTok && !eof()) {
1481         if (FormatTok->is(tok::r_brace)) {
1482           FormatTok->setFinalizedType(TT_InlineASMBrace);
1483           nextToken();
1484           addUnwrappedLine();
1485           break;
1486         }
1487         FormatTok->Finalized = true;
1488         nextToken();
1489       }
1490     }
1491     break;
1492   case tok::kw_namespace:
1493     parseNamespace();
1494     return;
1495   case tok::kw_if: {
1496     if (Style.isJavaScript() && Line->MustBeDeclaration) {
1497       // field/method declaration.
1498       break;
1499     }
1500     FormatToken *Tok = parseIfThenElse(IfKind);
1501     if (IfLeftBrace)
1502       *IfLeftBrace = Tok;
1503     return;
1504   }
1505   case tok::kw_for:
1506   case tok::kw_while:
1507     if (Style.isJavaScript() && Line->MustBeDeclaration) {
1508       // field/method declaration.
1509       break;
1510     }
1511     parseForOrWhileLoop();
1512     return;
1513   case tok::kw_do:
1514     if (Style.isJavaScript() && Line->MustBeDeclaration) {
1515       // field/method declaration.
1516       break;
1517     }
1518     parseDoWhile();
1519     if (HasDoWhile)
1520       *HasDoWhile = true;
1521     return;
1522   case tok::kw_switch:
1523     if (Style.isJavaScript() && Line->MustBeDeclaration) {
1524       // 'switch: string' field declaration.
1525       break;
1526     }
1527     parseSwitch(/*IsExpr=*/false);
1528     return;
1529   case tok::kw_default: {
1530     // In Verilog default along with other labels are handled in the next loop.
1531     if (Style.isVerilog())
1532       break;
1533     if (Style.isJavaScript() && Line->MustBeDeclaration) {
1534       // 'default: string' field declaration.
1535       break;
1536     }
1537     auto *Default = FormatTok;
1538     nextToken();
1539     if (FormatTok->is(tok::colon)) {
1540       FormatTok->setFinalizedType(TT_CaseLabelColon);
1541       parseLabel();
1542       return;
1543     }
1544     if (FormatTok->is(tok::arrow)) {
1545       FormatTok->setFinalizedType(TT_CaseLabelArrow);
1546       Default->setFinalizedType(TT_SwitchExpressionLabel);
1547       parseLabel();
1548       return;
1549     }
1550     // e.g. "default void f() {}" in a Java interface.
1551     break;
1552   }
1553   case tok::kw_case:
1554     // Proto: there are no switch/case statements.
1555     if (Style.Language == FormatStyle::LK_Proto) {
1556       nextToken();
1557       return;
1558     }
1559     if (Style.isVerilog()) {
1560       parseBlock();
1561       addUnwrappedLine();
1562       return;
1563     }
1564     if (Style.isJavaScript() && Line->MustBeDeclaration) {
1565       // 'case: string' field declaration.
1566       nextToken();
1567       break;
1568     }
1569     parseCaseLabel();
1570     return;
1571   case tok::kw_try:
1572   case tok::kw___try:
1573     if (Style.isJavaScript() && Line->MustBeDeclaration) {
1574       // field/method declaration.
1575       break;
1576     }
1577     parseTryCatch();
1578     return;
1579   case tok::kw_extern:
1580     nextToken();
1581     if (Style.isVerilog()) {
1582       // In Verilog and extern module declaration looks like a start of module.
1583       // But there is no body and endmodule. So we handle it separately.
1584       if (Keywords.isVerilogHierarchy(*FormatTok)) {
1585         parseVerilogHierarchyHeader();
1586         return;
1587       }
1588     } else if (FormatTok->is(tok::string_literal)) {
1589       nextToken();
1590       if (FormatTok->is(tok::l_brace)) {
1591         if (Style.BraceWrapping.AfterExternBlock)
1592           addUnwrappedLine();
1593         // Either we indent or for backwards compatibility we follow the
1594         // AfterExternBlock style.
1595         unsigned AddLevels =
1596             (Style.IndentExternBlock == FormatStyle::IEBS_Indent) ||
1597                     (Style.BraceWrapping.AfterExternBlock &&
1598                      Style.IndentExternBlock ==
1599                          FormatStyle::IEBS_AfterExternBlock)
1600                 ? 1u
1601                 : 0u;
1602         parseBlock(/*MustBeDeclaration=*/true, AddLevels);
1603         addUnwrappedLine();
1604         return;
1605       }
1606     }
1607     break;
1608   case tok::kw_export:
1609     if (Style.isJavaScript()) {
1610       parseJavaScriptEs6ImportExport();
1611       return;
1612     }
1613     if (IsCpp) {
1614       nextToken();
1615       if (FormatTok->is(tok::kw_namespace)) {
1616         parseNamespace();
1617         return;
1618       }
1619       if (FormatTok->is(Keywords.kw_import) && parseModuleImport())
1620         return;
1621     }
1622     break;
1623   case tok::kw_inline:
1624     nextToken();
1625     if (FormatTok->is(tok::kw_namespace)) {
1626       parseNamespace();
1627       return;
1628     }
1629     break;
1630   case tok::identifier:
1631     if (FormatTok->is(TT_ForEachMacro)) {
1632       parseForOrWhileLoop();
1633       return;
1634     }
1635     if (FormatTok->is(TT_MacroBlockBegin)) {
1636       parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u,
1637                  /*MunchSemi=*/false);
1638       return;
1639     }
1640     if (FormatTok->is(Keywords.kw_import)) {
1641       if (Style.isJavaScript()) {
1642         parseJavaScriptEs6ImportExport();
1643         return;
1644       }
1645       if (Style.Language == FormatStyle::LK_Proto) {
1646         nextToken();
1647         if (FormatTok->is(tok::kw_public))
1648           nextToken();
1649         if (FormatTok->isNot(tok::string_literal))
1650           return;
1651         nextToken();
1652         if (FormatTok->is(tok::semi))
1653           nextToken();
1654         addUnwrappedLine();
1655         return;
1656       }
1657       if (IsCpp && parseModuleImport())
1658         return;
1659     }
1660     if (IsCpp && FormatTok->isOneOf(Keywords.kw_signals, Keywords.kw_qsignals,
1661                                     Keywords.kw_slots, Keywords.kw_qslots)) {
1662       nextToken();
1663       if (FormatTok->is(tok::colon)) {
1664         nextToken();
1665         addUnwrappedLine();
1666         return;
1667       }
1668     }
1669     if (IsCpp && FormatTok->is(TT_StatementMacro)) {
1670       parseStatementMacro();
1671       return;
1672     }
1673     if (IsCpp && FormatTok->is(TT_NamespaceMacro)) {
1674       parseNamespace();
1675       return;
1676     }
1677     // In Verilog labels can be any expression, so we don't do them here.
1678     // JS doesn't have macros, and within classes colons indicate fields, not
1679     // labels.
1680     // TableGen doesn't have labels.
1681     if (!Style.isJavaScript() && !Style.isVerilog() && !Style.isTableGen() &&
1682         Tokens->peekNextToken()->is(tok::colon) && !Line->MustBeDeclaration) {
1683       nextToken();
1684       if (!Line->InMacroBody || CurrentLines->size() > 1)
1685         Line->Tokens.begin()->Tok->MustBreakBefore = true;
1686       FormatTok->setFinalizedType(TT_GotoLabelColon);
1687       parseLabel(!Style.IndentGotoLabels);
1688       if (HasLabel)
1689         *HasLabel = true;
1690       return;
1691     }
1692     // In all other cases, parse the declaration.
1693     break;
1694   default:
1695     break;
1696   }
1697 
1698   for (const bool InRequiresExpression =
1699            OpeningBrace && OpeningBrace->is(TT_RequiresExpressionLBrace);
1700        !eof();) {
1701     if (IsCpp && FormatTok->isCppAlternativeOperatorKeyword()) {
1702       if (auto *Next = Tokens->peekNextToken(/*SkipComment=*/true);
1703           Next && Next->isBinaryOperator()) {
1704         FormatTok->Tok.setKind(tok::identifier);
1705       }
1706     }
1707     const FormatToken *Previous = FormatTok->Previous;
1708     switch (FormatTok->Tok.getKind()) {
1709     case tok::at:
1710       nextToken();
1711       if (FormatTok->is(tok::l_brace)) {
1712         nextToken();
1713         parseBracedList();
1714         break;
1715       } else if (Style.Language == FormatStyle::LK_Java &&
1716                  FormatTok->is(Keywords.kw_interface)) {
1717         nextToken();
1718         break;
1719       }
1720       switch (FormatTok->Tok.getObjCKeywordID()) {
1721       case tok::objc_public:
1722       case tok::objc_protected:
1723       case tok::objc_package:
1724       case tok::objc_private:
1725         return parseAccessSpecifier();
1726       case tok::objc_interface:
1727       case tok::objc_implementation:
1728         return parseObjCInterfaceOrImplementation();
1729       case tok::objc_protocol:
1730         if (parseObjCProtocol())
1731           return;
1732         break;
1733       case tok::objc_end:
1734         return; // Handled by the caller.
1735       case tok::objc_optional:
1736       case tok::objc_required:
1737         nextToken();
1738         addUnwrappedLine();
1739         return;
1740       case tok::objc_autoreleasepool:
1741         nextToken();
1742         if (FormatTok->is(tok::l_brace)) {
1743           if (Style.BraceWrapping.AfterControlStatement ==
1744               FormatStyle::BWACS_Always) {
1745             addUnwrappedLine();
1746           }
1747           parseBlock();
1748         }
1749         addUnwrappedLine();
1750         return;
1751       case tok::objc_synchronized:
1752         nextToken();
1753         if (FormatTok->is(tok::l_paren)) {
1754           // Skip synchronization object
1755           parseParens();
1756         }
1757         if (FormatTok->is(tok::l_brace)) {
1758           if (Style.BraceWrapping.AfterControlStatement ==
1759               FormatStyle::BWACS_Always) {
1760             addUnwrappedLine();
1761           }
1762           parseBlock();
1763         }
1764         addUnwrappedLine();
1765         return;
1766       case tok::objc_try:
1767         // This branch isn't strictly necessary (the kw_try case below would
1768         // do this too after the tok::at is parsed above).  But be explicit.
1769         parseTryCatch();
1770         return;
1771       default:
1772         break;
1773       }
1774       break;
1775     case tok::kw_requires: {
1776       if (IsCpp) {
1777         bool ParsedClause = parseRequires();
1778         if (ParsedClause)
1779           return;
1780       } else {
1781         nextToken();
1782       }
1783       break;
1784     }
1785     case tok::kw_enum:
1786       // Ignore if this is part of "template <enum ..." or "... -> enum" or
1787       // "template <..., enum ...>".
1788       if (Previous && Previous->isOneOf(tok::less, tok::arrow, tok::comma)) {
1789         nextToken();
1790         break;
1791       }
1792 
1793       // parseEnum falls through and does not yet add an unwrapped line as an
1794       // enum definition can start a structural element.
1795       if (!parseEnum())
1796         break;
1797       // This only applies to C++ and Verilog.
1798       if (!IsCpp && !Style.isVerilog()) {
1799         addUnwrappedLine();
1800         return;
1801       }
1802       break;
1803     case tok::kw_typedef:
1804       nextToken();
1805       if (FormatTok->isOneOf(Keywords.kw_NS_ENUM, Keywords.kw_NS_OPTIONS,
1806                              Keywords.kw_CF_ENUM, Keywords.kw_CF_OPTIONS,
1807                              Keywords.kw_CF_CLOSED_ENUM,
1808                              Keywords.kw_NS_CLOSED_ENUM)) {
1809         parseEnum();
1810       }
1811       break;
1812     case tok::kw_class:
1813       if (Style.isVerilog()) {
1814         parseBlock();
1815         addUnwrappedLine();
1816         return;
1817       }
1818       if (Style.isTableGen()) {
1819         // Do nothing special. In this case the l_brace becomes FunctionLBrace.
1820         // This is same as def and so on.
1821         nextToken();
1822         break;
1823       }
1824       [[fallthrough]];
1825     case tok::kw_struct:
1826     case tok::kw_union:
1827       if (parseStructLike())
1828         return;
1829       break;
1830     case tok::kw_decltype:
1831       nextToken();
1832       if (FormatTok->is(tok::l_paren)) {
1833         parseParens();
1834         assert(FormatTok->Previous);
1835         if (FormatTok->Previous->endsSequence(tok::r_paren, tok::kw_auto,
1836                                               tok::l_paren)) {
1837           Line->SeenDecltypeAuto = true;
1838         }
1839       }
1840       break;
1841     case tok::period:
1842       nextToken();
1843       // In Java, classes have an implicit static member "class".
1844       if (Style.Language == FormatStyle::LK_Java && FormatTok &&
1845           FormatTok->is(tok::kw_class)) {
1846         nextToken();
1847       }
1848       if (Style.isJavaScript() && FormatTok &&
1849           FormatTok->Tok.getIdentifierInfo()) {
1850         // JavaScript only has pseudo keywords, all keywords are allowed to
1851         // appear in "IdentifierName" positions. See http://es5.github.io/#x7.6
1852         nextToken();
1853       }
1854       break;
1855     case tok::semi:
1856       nextToken();
1857       addUnwrappedLine();
1858       return;
1859     case tok::r_brace:
1860       addUnwrappedLine();
1861       return;
1862     case tok::l_paren: {
1863       parseParens();
1864       // Break the unwrapped line if a K&R C function definition has a parameter
1865       // declaration.
1866       if (OpeningBrace || !IsCpp || !Previous || eof())
1867         break;
1868       if (isC78ParameterDecl(FormatTok,
1869                              Tokens->peekNextToken(/*SkipComment=*/true),
1870                              Previous)) {
1871         addUnwrappedLine();
1872         return;
1873       }
1874       break;
1875     }
1876     case tok::kw_operator:
1877       nextToken();
1878       if (FormatTok->isBinaryOperator())
1879         nextToken();
1880       break;
1881     case tok::caret:
1882       nextToken();
1883       // Block return type.
1884       if (FormatTok->Tok.isAnyIdentifier() || FormatTok->isTypeName(LangOpts)) {
1885         nextToken();
1886         // Return types: pointers are ok too.
1887         while (FormatTok->is(tok::star))
1888           nextToken();
1889       }
1890       // Block argument list.
1891       if (FormatTok->is(tok::l_paren))
1892         parseParens();
1893       // Block body.
1894       if (FormatTok->is(tok::l_brace))
1895         parseChildBlock();
1896       break;
1897     case tok::l_brace:
1898       if (InRequiresExpression)
1899         FormatTok->setFinalizedType(TT_BracedListLBrace);
1900       if (!tryToParsePropertyAccessor() && !tryToParseBracedList()) {
1901         IsDecltypeAutoFunction = Line->SeenDecltypeAuto;
1902         // A block outside of parentheses must be the last part of a
1903         // structural element.
1904         // FIXME: Figure out cases where this is not true, and add projections
1905         // for them (the one we know is missing are lambdas).
1906         if (Style.Language == FormatStyle::LK_Java &&
1907             Line->Tokens.front().Tok->is(Keywords.kw_synchronized)) {
1908           // If necessary, we could set the type to something different than
1909           // TT_FunctionLBrace.
1910           if (Style.BraceWrapping.AfterControlStatement ==
1911               FormatStyle::BWACS_Always) {
1912             addUnwrappedLine();
1913           }
1914         } else if (Style.BraceWrapping.AfterFunction) {
1915           addUnwrappedLine();
1916         }
1917         if (!Previous || Previous->isNot(TT_TypeDeclarationParen))
1918           FormatTok->setFinalizedType(TT_FunctionLBrace);
1919         parseBlock();
1920         IsDecltypeAutoFunction = false;
1921         addUnwrappedLine();
1922         return;
1923       }
1924       // Otherwise this was a braced init list, and the structural
1925       // element continues.
1926       break;
1927     case tok::kw_try:
1928       if (Style.isJavaScript() && Line->MustBeDeclaration) {
1929         // field/method declaration.
1930         nextToken();
1931         break;
1932       }
1933       // We arrive here when parsing function-try blocks.
1934       if (Style.BraceWrapping.AfterFunction)
1935         addUnwrappedLine();
1936       parseTryCatch();
1937       return;
1938     case tok::identifier: {
1939       if (Style.isCSharp() && FormatTok->is(Keywords.kw_where) &&
1940           Line->MustBeDeclaration) {
1941         addUnwrappedLine();
1942         parseCSharpGenericTypeConstraint();
1943         break;
1944       }
1945       if (FormatTok->is(TT_MacroBlockEnd)) {
1946         addUnwrappedLine();
1947         return;
1948       }
1949 
1950       // Function declarations (as opposed to function expressions) are parsed
1951       // on their own unwrapped line by continuing this loop. Function
1952       // expressions (functions that are not on their own line) must not create
1953       // a new unwrapped line, so they are special cased below.
1954       size_t TokenCount = Line->Tokens.size();
1955       if (Style.isJavaScript() && FormatTok->is(Keywords.kw_function) &&
1956           (TokenCount > 1 ||
1957            (TokenCount == 1 &&
1958             Line->Tokens.front().Tok->isNot(Keywords.kw_async)))) {
1959         tryToParseJSFunction();
1960         break;
1961       }
1962       if ((Style.isJavaScript() || Style.Language == FormatStyle::LK_Java) &&
1963           FormatTok->is(Keywords.kw_interface)) {
1964         if (Style.isJavaScript()) {
1965           // In JavaScript/TypeScript, "interface" can be used as a standalone
1966           // identifier, e.g. in `var interface = 1;`. If "interface" is
1967           // followed by another identifier, it is very like to be an actual
1968           // interface declaration.
1969           unsigned StoredPosition = Tokens->getPosition();
1970           FormatToken *Next = Tokens->getNextToken();
1971           FormatTok = Tokens->setPosition(StoredPosition);
1972           if (!mustBeJSIdent(Keywords, Next)) {
1973             nextToken();
1974             break;
1975           }
1976         }
1977         parseRecord();
1978         addUnwrappedLine();
1979         return;
1980       }
1981 
1982       if (Style.isVerilog()) {
1983         if (FormatTok->is(Keywords.kw_table)) {
1984           parseVerilogTable();
1985           return;
1986         }
1987         if (Keywords.isVerilogBegin(*FormatTok) ||
1988             Keywords.isVerilogHierarchy(*FormatTok)) {
1989           parseBlock();
1990           addUnwrappedLine();
1991           return;
1992         }
1993       }
1994 
1995       if (!IsCpp && FormatTok->is(Keywords.kw_interface)) {
1996         if (parseStructLike())
1997           return;
1998         break;
1999       }
2000 
2001       if (IsCpp && FormatTok->is(TT_StatementMacro)) {
2002         parseStatementMacro();
2003         return;
2004       }
2005 
2006       // See if the following token should start a new unwrapped line.
2007       StringRef Text = FormatTok->TokenText;
2008 
2009       FormatToken *PreviousToken = FormatTok;
2010       nextToken();
2011 
2012       // JS doesn't have macros, and within classes colons indicate fields, not
2013       // labels.
2014       if (Style.isJavaScript())
2015         break;
2016 
2017       auto OneTokenSoFar = [&]() {
2018         auto I = Line->Tokens.begin(), E = Line->Tokens.end();
2019         while (I != E && I->Tok->is(tok::comment))
2020           ++I;
2021         if (Style.isVerilog())
2022           while (I != E && I->Tok->is(tok::hash))
2023             ++I;
2024         return I != E && (++I == E);
2025       };
2026       if (OneTokenSoFar()) {
2027         // Recognize function-like macro usages without trailing semicolon as
2028         // well as free-standing macros like Q_OBJECT.
2029         bool FunctionLike = FormatTok->is(tok::l_paren);
2030         if (FunctionLike)
2031           parseParens();
2032 
2033         bool FollowedByNewline =
2034             CommentsBeforeNextToken.empty()
2035                 ? FormatTok->NewlinesBefore > 0
2036                 : CommentsBeforeNextToken.front()->NewlinesBefore > 0;
2037 
2038         if (FollowedByNewline && (Text.size() >= 5 || FunctionLike) &&
2039             tokenCanStartNewLine(*FormatTok) && Text == Text.upper()) {
2040           if (PreviousToken->isNot(TT_UntouchableMacroFunc))
2041             PreviousToken->setFinalizedType(TT_FunctionLikeOrFreestandingMacro);
2042           addUnwrappedLine();
2043           return;
2044         }
2045       }
2046       break;
2047     }
2048     case tok::equal:
2049       if ((Style.isJavaScript() || Style.isCSharp()) &&
2050           FormatTok->is(TT_FatArrow)) {
2051         tryToParseChildBlock();
2052         break;
2053       }
2054 
2055       nextToken();
2056       if (FormatTok->is(tok::l_brace)) {
2057         // Block kind should probably be set to BK_BracedInit for any language.
2058         // C# needs this change to ensure that array initialisers and object
2059         // initialisers are indented the same way.
2060         if (Style.isCSharp())
2061           FormatTok->setBlockKind(BK_BracedInit);
2062         // TableGen's defset statement has syntax of the form,
2063         // `defset <type> <name> = { <statement>... }`
2064         if (Style.isTableGen() &&
2065             Line->Tokens.begin()->Tok->is(Keywords.kw_defset)) {
2066           FormatTok->setFinalizedType(TT_FunctionLBrace);
2067           parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u,
2068                      /*MunchSemi=*/false);
2069           addUnwrappedLine();
2070           break;
2071         }
2072         nextToken();
2073         parseBracedList();
2074       } else if (Style.Language == FormatStyle::LK_Proto &&
2075                  FormatTok->is(tok::less)) {
2076         nextToken();
2077         parseBracedList(/*IsAngleBracket=*/true);
2078       }
2079       break;
2080     case tok::l_square:
2081       parseSquare();
2082       break;
2083     case tok::kw_new:
2084       parseNew();
2085       break;
2086     case tok::kw_switch:
2087       if (Style.Language == FormatStyle::LK_Java)
2088         parseSwitch(/*IsExpr=*/true);
2089       nextToken();
2090       break;
2091     case tok::kw_case:
2092       // Proto: there are no switch/case statements.
2093       if (Style.Language == FormatStyle::LK_Proto) {
2094         nextToken();
2095         return;
2096       }
2097       // In Verilog switch is called case.
2098       if (Style.isVerilog()) {
2099         parseBlock();
2100         addUnwrappedLine();
2101         return;
2102       }
2103       if (Style.isJavaScript() && Line->MustBeDeclaration) {
2104         // 'case: string' field declaration.
2105         nextToken();
2106         break;
2107       }
2108       parseCaseLabel();
2109       break;
2110     case tok::kw_default:
2111       nextToken();
2112       if (Style.isVerilog()) {
2113         if (FormatTok->is(tok::colon)) {
2114           // The label will be handled in the next iteration.
2115           break;
2116         }
2117         if (FormatTok->is(Keywords.kw_clocking)) {
2118           // A default clocking block.
2119           parseBlock();
2120           addUnwrappedLine();
2121           return;
2122         }
2123         parseVerilogCaseLabel();
2124         return;
2125       }
2126       break;
2127     case tok::colon:
2128       nextToken();
2129       if (Style.isVerilog()) {
2130         parseVerilogCaseLabel();
2131         return;
2132       }
2133       break;
2134     case tok::greater:
2135       nextToken();
2136       if (FormatTok->is(tok::l_brace))
2137         FormatTok->Previous->setFinalizedType(TT_TemplateCloser);
2138       break;
2139     default:
2140       nextToken();
2141       break;
2142     }
2143   }
2144 }
2145 
2146 bool UnwrappedLineParser::tryToParsePropertyAccessor() {
2147   assert(FormatTok->is(tok::l_brace));
2148   if (!Style.isCSharp())
2149     return false;
2150   // See if it's a property accessor.
2151   if (FormatTok->Previous->isNot(tok::identifier))
2152     return false;
2153 
2154   // See if we are inside a property accessor.
2155   //
2156   // Record the current tokenPosition so that we can advance and
2157   // reset the current token. `Next` is not set yet so we need
2158   // another way to advance along the token stream.
2159   unsigned int StoredPosition = Tokens->getPosition();
2160   FormatToken *Tok = Tokens->getNextToken();
2161 
2162   // A trivial property accessor is of the form:
2163   // { [ACCESS_SPECIFIER] [get]; [ACCESS_SPECIFIER] [set|init] }
2164   // Track these as they do not require line breaks to be introduced.
2165   bool HasSpecialAccessor = false;
2166   bool IsTrivialPropertyAccessor = true;
2167   while (!eof()) {
2168     if (Tok->isAccessSpecifierKeyword() ||
2169         Tok->isOneOf(tok::semi, Keywords.kw_internal, Keywords.kw_get,
2170                      Keywords.kw_init, Keywords.kw_set)) {
2171       if (Tok->isOneOf(Keywords.kw_get, Keywords.kw_init, Keywords.kw_set))
2172         HasSpecialAccessor = true;
2173       Tok = Tokens->getNextToken();
2174       continue;
2175     }
2176     if (Tok->isNot(tok::r_brace))
2177       IsTrivialPropertyAccessor = false;
2178     break;
2179   }
2180 
2181   if (!HasSpecialAccessor) {
2182     Tokens->setPosition(StoredPosition);
2183     return false;
2184   }
2185 
2186   // Try to parse the property accessor:
2187   // https://docs.microsoft.com/en-us/dotnet/csharp/programming-guide/classes-and-structs/properties
2188   Tokens->setPosition(StoredPosition);
2189   if (!IsTrivialPropertyAccessor && Style.BraceWrapping.AfterFunction)
2190     addUnwrappedLine();
2191   nextToken();
2192   do {
2193     switch (FormatTok->Tok.getKind()) {
2194     case tok::r_brace:
2195       nextToken();
2196       if (FormatTok->is(tok::equal)) {
2197         while (!eof() && FormatTok->isNot(tok::semi))
2198           nextToken();
2199         nextToken();
2200       }
2201       addUnwrappedLine();
2202       return true;
2203     case tok::l_brace:
2204       ++Line->Level;
2205       parseBlock(/*MustBeDeclaration=*/true);
2206       addUnwrappedLine();
2207       --Line->Level;
2208       break;
2209     case tok::equal:
2210       if (FormatTok->is(TT_FatArrow)) {
2211         ++Line->Level;
2212         do {
2213           nextToken();
2214         } while (!eof() && FormatTok->isNot(tok::semi));
2215         nextToken();
2216         addUnwrappedLine();
2217         --Line->Level;
2218         break;
2219       }
2220       nextToken();
2221       break;
2222     default:
2223       if (FormatTok->isOneOf(Keywords.kw_get, Keywords.kw_init,
2224                              Keywords.kw_set) &&
2225           !IsTrivialPropertyAccessor) {
2226         // Non-trivial get/set needs to be on its own line.
2227         addUnwrappedLine();
2228       }
2229       nextToken();
2230     }
2231   } while (!eof());
2232 
2233   // Unreachable for well-formed code (paired '{' and '}').
2234   return true;
2235 }
2236 
2237 bool UnwrappedLineParser::tryToParseLambda() {
2238   assert(FormatTok->is(tok::l_square));
2239   if (!IsCpp) {
2240     nextToken();
2241     return false;
2242   }
2243   FormatToken &LSquare = *FormatTok;
2244   if (!tryToParseLambdaIntroducer())
2245     return false;
2246 
2247   bool SeenArrow = false;
2248   bool InTemplateParameterList = false;
2249 
2250   while (FormatTok->isNot(tok::l_brace)) {
2251     if (FormatTok->isTypeName(LangOpts) || FormatTok->isAttribute()) {
2252       nextToken();
2253       continue;
2254     }
2255     switch (FormatTok->Tok.getKind()) {
2256     case tok::l_brace:
2257       break;
2258     case tok::l_paren:
2259       parseParens(/*AmpAmpTokenType=*/TT_PointerOrReference);
2260       break;
2261     case tok::l_square:
2262       parseSquare();
2263       break;
2264     case tok::less:
2265       assert(FormatTok->Previous);
2266       if (FormatTok->Previous->is(tok::r_square))
2267         InTemplateParameterList = true;
2268       nextToken();
2269       break;
2270     case tok::kw_auto:
2271     case tok::kw_class:
2272     case tok::kw_struct:
2273     case tok::kw_union:
2274     case tok::kw_template:
2275     case tok::kw_typename:
2276     case tok::amp:
2277     case tok::star:
2278     case tok::kw_const:
2279     case tok::kw_constexpr:
2280     case tok::kw_consteval:
2281     case tok::comma:
2282     case tok::greater:
2283     case tok::identifier:
2284     case tok::numeric_constant:
2285     case tok::coloncolon:
2286     case tok::kw_mutable:
2287     case tok::kw_noexcept:
2288     case tok::kw_static:
2289       nextToken();
2290       break;
2291     // Specialization of a template with an integer parameter can contain
2292     // arithmetic, logical, comparison and ternary operators.
2293     //
2294     // FIXME: This also accepts sequences of operators that are not in the scope
2295     // of a template argument list.
2296     //
2297     // In a C++ lambda a template type can only occur after an arrow. We use
2298     // this as an heuristic to distinguish between Objective-C expressions
2299     // followed by an `a->b` expression, such as:
2300     // ([obj func:arg] + a->b)
2301     // Otherwise the code below would parse as a lambda.
2302     case tok::plus:
2303     case tok::minus:
2304     case tok::exclaim:
2305     case tok::tilde:
2306     case tok::slash:
2307     case tok::percent:
2308     case tok::lessless:
2309     case tok::pipe:
2310     case tok::pipepipe:
2311     case tok::ampamp:
2312     case tok::caret:
2313     case tok::equalequal:
2314     case tok::exclaimequal:
2315     case tok::greaterequal:
2316     case tok::lessequal:
2317     case tok::question:
2318     case tok::colon:
2319     case tok::ellipsis:
2320     case tok::kw_true:
2321     case tok::kw_false:
2322       if (SeenArrow || InTemplateParameterList) {
2323         nextToken();
2324         break;
2325       }
2326       return true;
2327     case tok::arrow:
2328       // This might or might not actually be a lambda arrow (this could be an
2329       // ObjC method invocation followed by a dereferencing arrow). We might
2330       // reset this back to TT_Unknown in TokenAnnotator.
2331       FormatTok->setFinalizedType(TT_LambdaArrow);
2332       SeenArrow = true;
2333       nextToken();
2334       break;
2335     case tok::kw_requires: {
2336       auto *RequiresToken = FormatTok;
2337       nextToken();
2338       parseRequiresClause(RequiresToken);
2339       break;
2340     }
2341     case tok::equal:
2342       if (!InTemplateParameterList)
2343         return true;
2344       nextToken();
2345       break;
2346     default:
2347       return true;
2348     }
2349   }
2350 
2351   FormatTok->setFinalizedType(TT_LambdaLBrace);
2352   LSquare.setFinalizedType(TT_LambdaLSquare);
2353 
2354   NestedLambdas.push_back(Line->SeenDecltypeAuto);
2355   parseChildBlock();
2356   assert(!NestedLambdas.empty());
2357   NestedLambdas.pop_back();
2358 
2359   return true;
2360 }
2361 
2362 bool UnwrappedLineParser::tryToParseLambdaIntroducer() {
2363   const FormatToken *Previous = FormatTok->Previous;
2364   const FormatToken *LeftSquare = FormatTok;
2365   nextToken();
2366   if ((Previous && ((Previous->Tok.getIdentifierInfo() &&
2367                      !Previous->isOneOf(tok::kw_return, tok::kw_co_await,
2368                                         tok::kw_co_yield, tok::kw_co_return)) ||
2369                     Previous->closesScope())) ||
2370       LeftSquare->isCppStructuredBinding(IsCpp)) {
2371     return false;
2372   }
2373   if (FormatTok->is(tok::l_square) || tok::isLiteral(FormatTok->Tok.getKind()))
2374     return false;
2375   if (FormatTok->is(tok::r_square)) {
2376     const FormatToken *Next = Tokens->peekNextToken(/*SkipComment=*/true);
2377     if (Next->is(tok::greater))
2378       return false;
2379   }
2380   parseSquare(/*LambdaIntroducer=*/true);
2381   return true;
2382 }
2383 
2384 void UnwrappedLineParser::tryToParseJSFunction() {
2385   assert(FormatTok->is(Keywords.kw_function));
2386   if (FormatTok->is(Keywords.kw_async))
2387     nextToken();
2388   // Consume "function".
2389   nextToken();
2390 
2391   // Consume * (generator function). Treat it like C++'s overloaded operators.
2392   if (FormatTok->is(tok::star)) {
2393     FormatTok->setFinalizedType(TT_OverloadedOperator);
2394     nextToken();
2395   }
2396 
2397   // Consume function name.
2398   if (FormatTok->is(tok::identifier))
2399     nextToken();
2400 
2401   if (FormatTok->isNot(tok::l_paren))
2402     return;
2403 
2404   // Parse formal parameter list.
2405   parseParens();
2406 
2407   if (FormatTok->is(tok::colon)) {
2408     // Parse a type definition.
2409     nextToken();
2410 
2411     // Eat the type declaration. For braced inline object types, balance braces,
2412     // otherwise just parse until finding an l_brace for the function body.
2413     if (FormatTok->is(tok::l_brace))
2414       tryToParseBracedList();
2415     else
2416       while (!FormatTok->isOneOf(tok::l_brace, tok::semi) && !eof())
2417         nextToken();
2418   }
2419 
2420   if (FormatTok->is(tok::semi))
2421     return;
2422 
2423   parseChildBlock();
2424 }
2425 
2426 bool UnwrappedLineParser::tryToParseBracedList() {
2427   if (FormatTok->is(BK_Unknown))
2428     calculateBraceTypes();
2429   assert(FormatTok->isNot(BK_Unknown));
2430   if (FormatTok->is(BK_Block))
2431     return false;
2432   nextToken();
2433   parseBracedList();
2434   return true;
2435 }
2436 
2437 bool UnwrappedLineParser::tryToParseChildBlock() {
2438   assert(Style.isJavaScript() || Style.isCSharp());
2439   assert(FormatTok->is(TT_FatArrow));
2440   // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType TT_FatArrow.
2441   // They always start an expression or a child block if followed by a curly
2442   // brace.
2443   nextToken();
2444   if (FormatTok->isNot(tok::l_brace))
2445     return false;
2446   parseChildBlock();
2447   return true;
2448 }
2449 
2450 bool UnwrappedLineParser::parseBracedList(bool IsAngleBracket, bool IsEnum) {
2451   assert(!IsAngleBracket || !IsEnum);
2452   bool HasError = false;
2453 
2454   // FIXME: Once we have an expression parser in the UnwrappedLineParser,
2455   // replace this by using parseAssignmentExpression() inside.
2456   do {
2457     if (Style.isCSharp() && FormatTok->is(TT_FatArrow) &&
2458         tryToParseChildBlock()) {
2459       continue;
2460     }
2461     if (Style.isJavaScript()) {
2462       if (FormatTok->is(Keywords.kw_function)) {
2463         tryToParseJSFunction();
2464         continue;
2465       }
2466       if (FormatTok->is(tok::l_brace)) {
2467         // Could be a method inside of a braced list `{a() { return 1; }}`.
2468         if (tryToParseBracedList())
2469           continue;
2470         parseChildBlock();
2471       }
2472     }
2473     if (FormatTok->is(IsAngleBracket ? tok::greater : tok::r_brace)) {
2474       if (IsEnum) {
2475         FormatTok->setBlockKind(BK_Block);
2476         if (!Style.AllowShortEnumsOnASingleLine)
2477           addUnwrappedLine();
2478       }
2479       nextToken();
2480       return !HasError;
2481     }
2482     switch (FormatTok->Tok.getKind()) {
2483     case tok::l_square:
2484       if (Style.isCSharp())
2485         parseSquare();
2486       else
2487         tryToParseLambda();
2488       break;
2489     case tok::l_paren:
2490       parseParens();
2491       // JavaScript can just have free standing methods and getters/setters in
2492       // object literals. Detect them by a "{" following ")".
2493       if (Style.isJavaScript()) {
2494         if (FormatTok->is(tok::l_brace))
2495           parseChildBlock();
2496         break;
2497       }
2498       break;
2499     case tok::l_brace:
2500       // Assume there are no blocks inside a braced init list apart
2501       // from the ones we explicitly parse out (like lambdas).
2502       FormatTok->setBlockKind(BK_BracedInit);
2503       if (!IsAngleBracket) {
2504         auto *Prev = FormatTok->Previous;
2505         if (Prev && Prev->is(tok::greater))
2506           Prev->setFinalizedType(TT_TemplateCloser);
2507       }
2508       nextToken();
2509       parseBracedList();
2510       break;
2511     case tok::less:
2512       nextToken();
2513       if (IsAngleBracket)
2514         parseBracedList(/*IsAngleBracket=*/true);
2515       break;
2516     case tok::semi:
2517       // JavaScript (or more precisely TypeScript) can have semicolons in braced
2518       // lists (in so-called TypeMemberLists). Thus, the semicolon cannot be
2519       // used for error recovery if we have otherwise determined that this is
2520       // a braced list.
2521       if (Style.isJavaScript()) {
2522         nextToken();
2523         break;
2524       }
2525       HasError = true;
2526       if (!IsEnum)
2527         return false;
2528       nextToken();
2529       break;
2530     case tok::comma:
2531       nextToken();
2532       if (IsEnum && !Style.AllowShortEnumsOnASingleLine)
2533         addUnwrappedLine();
2534       break;
2535     default:
2536       nextToken();
2537       break;
2538     }
2539   } while (!eof());
2540   return false;
2541 }
2542 
2543 /// \brief Parses a pair of parentheses (and everything between them).
2544 /// \param AmpAmpTokenType If different than TT_Unknown sets this type for all
2545 /// double ampersands. This applies for all nested scopes as well.
2546 ///
2547 /// Returns whether there is a `=` token between the parentheses.
2548 bool UnwrappedLineParser::parseParens(TokenType AmpAmpTokenType) {
2549   assert(FormatTok->is(tok::l_paren) && "'(' expected.");
2550   auto *LeftParen = FormatTok;
2551   bool SeenEqual = false;
2552   bool MightBeFoldExpr = false;
2553   const bool MightBeStmtExpr = Tokens->peekNextToken()->is(tok::l_brace);
2554   nextToken();
2555   do {
2556     switch (FormatTok->Tok.getKind()) {
2557     case tok::l_paren:
2558       if (parseParens(AmpAmpTokenType))
2559         SeenEqual = true;
2560       if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_brace))
2561         parseChildBlock();
2562       break;
2563     case tok::r_paren: {
2564       auto *Prev = LeftParen->Previous;
2565       if (!MightBeStmtExpr && !MightBeFoldExpr && !Line->InMacroBody &&
2566           Style.RemoveParentheses > FormatStyle::RPS_Leave) {
2567         const auto *Next = Tokens->peekNextToken();
2568         const bool DoubleParens =
2569             Prev && Prev->is(tok::l_paren) && Next && Next->is(tok::r_paren);
2570         const auto *PrevPrev = Prev ? Prev->getPreviousNonComment() : nullptr;
2571         const bool Blacklisted =
2572             PrevPrev &&
2573             (PrevPrev->isOneOf(tok::kw___attribute, tok::kw_decltype) ||
2574              (SeenEqual &&
2575               (PrevPrev->isOneOf(tok::kw_if, tok::kw_while) ||
2576                PrevPrev->endsSequence(tok::kw_constexpr, tok::kw_if))));
2577         const bool ReturnParens =
2578             Style.RemoveParentheses == FormatStyle::RPS_ReturnStatement &&
2579             ((NestedLambdas.empty() && !IsDecltypeAutoFunction) ||
2580              (!NestedLambdas.empty() && !NestedLambdas.back())) &&
2581             Prev && Prev->isOneOf(tok::kw_return, tok::kw_co_return) && Next &&
2582             Next->is(tok::semi);
2583         if ((DoubleParens && !Blacklisted) || ReturnParens) {
2584           LeftParen->Optional = true;
2585           FormatTok->Optional = true;
2586         }
2587       }
2588       if (Prev) {
2589         if (Prev->is(TT_TypenameMacro)) {
2590           LeftParen->setFinalizedType(TT_TypeDeclarationParen);
2591           FormatTok->setFinalizedType(TT_TypeDeclarationParen);
2592         } else if (Prev->is(tok::greater) && FormatTok->Previous == LeftParen) {
2593           Prev->setFinalizedType(TT_TemplateCloser);
2594         }
2595       }
2596       nextToken();
2597       return SeenEqual;
2598     }
2599     case tok::r_brace:
2600       // A "}" inside parenthesis is an error if there wasn't a matching "{".
2601       return SeenEqual;
2602     case tok::l_square:
2603       tryToParseLambda();
2604       break;
2605     case tok::l_brace:
2606       if (!tryToParseBracedList())
2607         parseChildBlock();
2608       break;
2609     case tok::at:
2610       nextToken();
2611       if (FormatTok->is(tok::l_brace)) {
2612         nextToken();
2613         parseBracedList();
2614       }
2615       break;
2616     case tok::ellipsis:
2617       MightBeFoldExpr = true;
2618       nextToken();
2619       break;
2620     case tok::equal:
2621       SeenEqual = true;
2622       if (Style.isCSharp() && FormatTok->is(TT_FatArrow))
2623         tryToParseChildBlock();
2624       else
2625         nextToken();
2626       break;
2627     case tok::kw_class:
2628       if (Style.isJavaScript())
2629         parseRecord(/*ParseAsExpr=*/true);
2630       else
2631         nextToken();
2632       break;
2633     case tok::identifier:
2634       if (Style.isJavaScript() && (FormatTok->is(Keywords.kw_function)))
2635         tryToParseJSFunction();
2636       else
2637         nextToken();
2638       break;
2639     case tok::kw_switch:
2640       parseSwitch(/*IsExpr=*/true);
2641       break;
2642     case tok::kw_requires: {
2643       auto RequiresToken = FormatTok;
2644       nextToken();
2645       parseRequiresExpression(RequiresToken);
2646       break;
2647     }
2648     case tok::ampamp:
2649       if (AmpAmpTokenType != TT_Unknown)
2650         FormatTok->setFinalizedType(AmpAmpTokenType);
2651       [[fallthrough]];
2652     default:
2653       nextToken();
2654       break;
2655     }
2656   } while (!eof());
2657   return SeenEqual;
2658 }
2659 
2660 void UnwrappedLineParser::parseSquare(bool LambdaIntroducer) {
2661   if (!LambdaIntroducer) {
2662     assert(FormatTok->is(tok::l_square) && "'[' expected.");
2663     if (tryToParseLambda())
2664       return;
2665   }
2666   do {
2667     switch (FormatTok->Tok.getKind()) {
2668     case tok::l_paren:
2669       parseParens();
2670       break;
2671     case tok::r_square:
2672       nextToken();
2673       return;
2674     case tok::r_brace:
2675       // A "}" inside parenthesis is an error if there wasn't a matching "{".
2676       return;
2677     case tok::l_square:
2678       parseSquare();
2679       break;
2680     case tok::l_brace: {
2681       if (!tryToParseBracedList())
2682         parseChildBlock();
2683       break;
2684     }
2685     case tok::at:
2686     case tok::colon:
2687       nextToken();
2688       if (FormatTok->is(tok::l_brace)) {
2689         nextToken();
2690         parseBracedList();
2691       }
2692       break;
2693     default:
2694       nextToken();
2695       break;
2696     }
2697   } while (!eof());
2698 }
2699 
2700 void UnwrappedLineParser::keepAncestorBraces() {
2701   if (!Style.RemoveBracesLLVM)
2702     return;
2703 
2704   const int MaxNestingLevels = 2;
2705   const int Size = NestedTooDeep.size();
2706   if (Size >= MaxNestingLevels)
2707     NestedTooDeep[Size - MaxNestingLevels] = true;
2708   NestedTooDeep.push_back(false);
2709 }
2710 
2711 static FormatToken *getLastNonComment(const UnwrappedLine &Line) {
2712   for (const auto &Token : llvm::reverse(Line.Tokens))
2713     if (Token.Tok->isNot(tok::comment))
2714       return Token.Tok;
2715 
2716   return nullptr;
2717 }
2718 
2719 void UnwrappedLineParser::parseUnbracedBody(bool CheckEOF) {
2720   FormatToken *Tok = nullptr;
2721 
2722   if (Style.InsertBraces && !Line->InPPDirective && !Line->Tokens.empty() &&
2723       PreprocessorDirectives.empty() && FormatTok->isNot(tok::semi)) {
2724     Tok = Style.BraceWrapping.AfterControlStatement == FormatStyle::BWACS_Never
2725               ? getLastNonComment(*Line)
2726               : Line->Tokens.back().Tok;
2727     assert(Tok);
2728     if (Tok->BraceCount < 0) {
2729       assert(Tok->BraceCount == -1);
2730       Tok = nullptr;
2731     } else {
2732       Tok->BraceCount = -1;
2733     }
2734   }
2735 
2736   addUnwrappedLine();
2737   ++Line->Level;
2738   ++Line->UnbracedBodyLevel;
2739   parseStructuralElement();
2740   --Line->UnbracedBodyLevel;
2741 
2742   if (Tok) {
2743     assert(!Line->InPPDirective);
2744     Tok = nullptr;
2745     for (const auto &L : llvm::reverse(*CurrentLines)) {
2746       if (!L.InPPDirective && getLastNonComment(L)) {
2747         Tok = L.Tokens.back().Tok;
2748         break;
2749       }
2750     }
2751     assert(Tok);
2752     ++Tok->BraceCount;
2753   }
2754 
2755   if (CheckEOF && eof())
2756     addUnwrappedLine();
2757 
2758   --Line->Level;
2759 }
2760 
2761 static void markOptionalBraces(FormatToken *LeftBrace) {
2762   if (!LeftBrace)
2763     return;
2764 
2765   assert(LeftBrace->is(tok::l_brace));
2766 
2767   FormatToken *RightBrace = LeftBrace->MatchingParen;
2768   if (!RightBrace) {
2769     assert(!LeftBrace->Optional);
2770     return;
2771   }
2772 
2773   assert(RightBrace->is(tok::r_brace));
2774   assert(RightBrace->MatchingParen == LeftBrace);
2775   assert(LeftBrace->Optional == RightBrace->Optional);
2776 
2777   LeftBrace->Optional = true;
2778   RightBrace->Optional = true;
2779 }
2780 
2781 void UnwrappedLineParser::handleAttributes() {
2782   // Handle AttributeMacro, e.g. `if (x) UNLIKELY`.
2783   if (FormatTok->isAttribute())
2784     nextToken();
2785   else if (FormatTok->is(tok::l_square))
2786     handleCppAttributes();
2787 }
2788 
2789 bool UnwrappedLineParser::handleCppAttributes() {
2790   // Handle [[likely]] / [[unlikely]] attributes.
2791   assert(FormatTok->is(tok::l_square));
2792   if (!tryToParseSimpleAttribute())
2793     return false;
2794   parseSquare();
2795   return true;
2796 }
2797 
2798 /// Returns whether \c Tok begins a block.
2799 bool UnwrappedLineParser::isBlockBegin(const FormatToken &Tok) const {
2800   // FIXME: rename the function or make
2801   // Tok.isOneOf(tok::l_brace, TT_MacroBlockBegin) work.
2802   return Style.isVerilog() ? Keywords.isVerilogBegin(Tok)
2803                            : Tok.is(tok::l_brace);
2804 }
2805 
2806 FormatToken *UnwrappedLineParser::parseIfThenElse(IfStmtKind *IfKind,
2807                                                   bool KeepBraces,
2808                                                   bool IsVerilogAssert) {
2809   assert((FormatTok->is(tok::kw_if) ||
2810           (Style.isVerilog() &&
2811            FormatTok->isOneOf(tok::kw_restrict, Keywords.kw_assert,
2812                               Keywords.kw_assume, Keywords.kw_cover))) &&
2813          "'if' expected");
2814   nextToken();
2815 
2816   if (IsVerilogAssert) {
2817     // Handle `assert #0` and `assert final`.
2818     if (FormatTok->is(Keywords.kw_verilogHash)) {
2819       nextToken();
2820       if (FormatTok->is(tok::numeric_constant))
2821         nextToken();
2822     } else if (FormatTok->isOneOf(Keywords.kw_final, Keywords.kw_property,
2823                                   Keywords.kw_sequence)) {
2824       nextToken();
2825     }
2826   }
2827 
2828   // TableGen's if statement has the form of `if <cond> then { ... }`.
2829   if (Style.isTableGen()) {
2830     while (!eof() && FormatTok->isNot(Keywords.kw_then)) {
2831       // Simply skip until then. This range only contains a value.
2832       nextToken();
2833     }
2834   }
2835 
2836   // Handle `if !consteval`.
2837   if (FormatTok->is(tok::exclaim))
2838     nextToken();
2839 
2840   bool KeepIfBraces = true;
2841   if (FormatTok->is(tok::kw_consteval)) {
2842     nextToken();
2843   } else {
2844     KeepIfBraces = !Style.RemoveBracesLLVM || KeepBraces;
2845     if (FormatTok->isOneOf(tok::kw_constexpr, tok::identifier))
2846       nextToken();
2847     if (FormatTok->is(tok::l_paren)) {
2848       FormatTok->setFinalizedType(TT_ConditionLParen);
2849       parseParens();
2850     }
2851   }
2852   handleAttributes();
2853   // The then action is optional in Verilog assert statements.
2854   if (IsVerilogAssert && FormatTok->is(tok::semi)) {
2855     nextToken();
2856     addUnwrappedLine();
2857     return nullptr;
2858   }
2859 
2860   bool NeedsUnwrappedLine = false;
2861   keepAncestorBraces();
2862 
2863   FormatToken *IfLeftBrace = nullptr;
2864   IfStmtKind IfBlockKind = IfStmtKind::NotIf;
2865 
2866   if (isBlockBegin(*FormatTok)) {
2867     FormatTok->setFinalizedType(TT_ControlStatementLBrace);
2868     IfLeftBrace = FormatTok;
2869     CompoundStatementIndenter Indenter(this, Style, Line->Level);
2870     parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u,
2871                /*MunchSemi=*/true, KeepIfBraces, &IfBlockKind);
2872     setPreviousRBraceType(TT_ControlStatementRBrace);
2873     if (Style.BraceWrapping.BeforeElse)
2874       addUnwrappedLine();
2875     else
2876       NeedsUnwrappedLine = true;
2877   } else if (IsVerilogAssert && FormatTok->is(tok::kw_else)) {
2878     addUnwrappedLine();
2879   } else {
2880     parseUnbracedBody();
2881   }
2882 
2883   if (Style.RemoveBracesLLVM) {
2884     assert(!NestedTooDeep.empty());
2885     KeepIfBraces = KeepIfBraces ||
2886                    (IfLeftBrace && !IfLeftBrace->MatchingParen) ||
2887                    NestedTooDeep.back() || IfBlockKind == IfStmtKind::IfOnly ||
2888                    IfBlockKind == IfStmtKind::IfElseIf;
2889   }
2890 
2891   bool KeepElseBraces = KeepIfBraces;
2892   FormatToken *ElseLeftBrace = nullptr;
2893   IfStmtKind Kind = IfStmtKind::IfOnly;
2894 
2895   if (FormatTok->is(tok::kw_else)) {
2896     if (Style.RemoveBracesLLVM) {
2897       NestedTooDeep.back() = false;
2898       Kind = IfStmtKind::IfElse;
2899     }
2900     nextToken();
2901     handleAttributes();
2902     if (isBlockBegin(*FormatTok)) {
2903       const bool FollowedByIf = Tokens->peekNextToken()->is(tok::kw_if);
2904       FormatTok->setFinalizedType(TT_ElseLBrace);
2905       ElseLeftBrace = FormatTok;
2906       CompoundStatementIndenter Indenter(this, Style, Line->Level);
2907       IfStmtKind ElseBlockKind = IfStmtKind::NotIf;
2908       FormatToken *IfLBrace =
2909           parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u,
2910                      /*MunchSemi=*/true, KeepElseBraces, &ElseBlockKind);
2911       setPreviousRBraceType(TT_ElseRBrace);
2912       if (FormatTok->is(tok::kw_else)) {
2913         KeepElseBraces = KeepElseBraces ||
2914                          ElseBlockKind == IfStmtKind::IfOnly ||
2915                          ElseBlockKind == IfStmtKind::IfElseIf;
2916       } else if (FollowedByIf && IfLBrace && !IfLBrace->Optional) {
2917         KeepElseBraces = true;
2918         assert(ElseLeftBrace->MatchingParen);
2919         markOptionalBraces(ElseLeftBrace);
2920       }
2921       addUnwrappedLine();
2922     } else if (!IsVerilogAssert && FormatTok->is(tok::kw_if)) {
2923       const FormatToken *Previous = Tokens->getPreviousToken();
2924       assert(Previous);
2925       const bool IsPrecededByComment = Previous->is(tok::comment);
2926       if (IsPrecededByComment) {
2927         addUnwrappedLine();
2928         ++Line->Level;
2929       }
2930       bool TooDeep = true;
2931       if (Style.RemoveBracesLLVM) {
2932         Kind = IfStmtKind::IfElseIf;
2933         TooDeep = NestedTooDeep.pop_back_val();
2934       }
2935       ElseLeftBrace = parseIfThenElse(/*IfKind=*/nullptr, KeepIfBraces);
2936       if (Style.RemoveBracesLLVM)
2937         NestedTooDeep.push_back(TooDeep);
2938       if (IsPrecededByComment)
2939         --Line->Level;
2940     } else {
2941       parseUnbracedBody(/*CheckEOF=*/true);
2942     }
2943   } else {
2944     KeepIfBraces = KeepIfBraces || IfBlockKind == IfStmtKind::IfElse;
2945     if (NeedsUnwrappedLine)
2946       addUnwrappedLine();
2947   }
2948 
2949   if (!Style.RemoveBracesLLVM)
2950     return nullptr;
2951 
2952   assert(!NestedTooDeep.empty());
2953   KeepElseBraces = KeepElseBraces ||
2954                    (ElseLeftBrace && !ElseLeftBrace->MatchingParen) ||
2955                    NestedTooDeep.back();
2956 
2957   NestedTooDeep.pop_back();
2958 
2959   if (!KeepIfBraces && !KeepElseBraces) {
2960     markOptionalBraces(IfLeftBrace);
2961     markOptionalBraces(ElseLeftBrace);
2962   } else if (IfLeftBrace) {
2963     FormatToken *IfRightBrace = IfLeftBrace->MatchingParen;
2964     if (IfRightBrace) {
2965       assert(IfRightBrace->MatchingParen == IfLeftBrace);
2966       assert(!IfLeftBrace->Optional);
2967       assert(!IfRightBrace->Optional);
2968       IfLeftBrace->MatchingParen = nullptr;
2969       IfRightBrace->MatchingParen = nullptr;
2970     }
2971   }
2972 
2973   if (IfKind)
2974     *IfKind = Kind;
2975 
2976   return IfLeftBrace;
2977 }
2978 
2979 void UnwrappedLineParser::parseTryCatch() {
2980   assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected");
2981   nextToken();
2982   bool NeedsUnwrappedLine = false;
2983   bool HasCtorInitializer = false;
2984   if (FormatTok->is(tok::colon)) {
2985     auto *Colon = FormatTok;
2986     // We are in a function try block, what comes is an initializer list.
2987     nextToken();
2988     if (FormatTok->is(tok::identifier)) {
2989       HasCtorInitializer = true;
2990       Colon->setFinalizedType(TT_CtorInitializerColon);
2991     }
2992 
2993     // In case identifiers were removed by clang-tidy, what might follow is
2994     // multiple commas in sequence - before the first identifier.
2995     while (FormatTok->is(tok::comma))
2996       nextToken();
2997 
2998     while (FormatTok->is(tok::identifier)) {
2999       nextToken();
3000       if (FormatTok->is(tok::l_paren)) {
3001         parseParens();
3002       } else if (FormatTok->is(tok::l_brace)) {
3003         nextToken();
3004         parseBracedList();
3005       }
3006 
3007       // In case identifiers were removed by clang-tidy, what might follow is
3008       // multiple commas in sequence - after the first identifier.
3009       while (FormatTok->is(tok::comma))
3010         nextToken();
3011     }
3012   }
3013   // Parse try with resource.
3014   if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::l_paren))
3015     parseParens();
3016 
3017   keepAncestorBraces();
3018 
3019   if (FormatTok->is(tok::l_brace)) {
3020     if (HasCtorInitializer)
3021       FormatTok->setFinalizedType(TT_FunctionLBrace);
3022     CompoundStatementIndenter Indenter(this, Style, Line->Level);
3023     parseBlock();
3024     if (Style.BraceWrapping.BeforeCatch)
3025       addUnwrappedLine();
3026     else
3027       NeedsUnwrappedLine = true;
3028   } else if (FormatTok->isNot(tok::kw_catch)) {
3029     // The C++ standard requires a compound-statement after a try.
3030     // If there's none, we try to assume there's a structuralElement
3031     // and try to continue.
3032     addUnwrappedLine();
3033     ++Line->Level;
3034     parseStructuralElement();
3035     --Line->Level;
3036   }
3037   while (true) {
3038     if (FormatTok->is(tok::at))
3039       nextToken();
3040     if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except,
3041                              tok::kw___finally) ||
3042           ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) &&
3043            FormatTok->is(Keywords.kw_finally)) ||
3044           (FormatTok->isObjCAtKeyword(tok::objc_catch) ||
3045            FormatTok->isObjCAtKeyword(tok::objc_finally)))) {
3046       break;
3047     }
3048     nextToken();
3049     while (FormatTok->isNot(tok::l_brace)) {
3050       if (FormatTok->is(tok::l_paren)) {
3051         parseParens();
3052         continue;
3053       }
3054       if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof)) {
3055         if (Style.RemoveBracesLLVM)
3056           NestedTooDeep.pop_back();
3057         return;
3058       }
3059       nextToken();
3060     }
3061     NeedsUnwrappedLine = false;
3062     Line->MustBeDeclaration = false;
3063     CompoundStatementIndenter Indenter(this, Style, Line->Level);
3064     parseBlock();
3065     if (Style.BraceWrapping.BeforeCatch)
3066       addUnwrappedLine();
3067     else
3068       NeedsUnwrappedLine = true;
3069   }
3070 
3071   if (Style.RemoveBracesLLVM)
3072     NestedTooDeep.pop_back();
3073 
3074   if (NeedsUnwrappedLine)
3075     addUnwrappedLine();
3076 }
3077 
3078 void UnwrappedLineParser::parseNamespace() {
3079   assert(FormatTok->isOneOf(tok::kw_namespace, TT_NamespaceMacro) &&
3080          "'namespace' expected");
3081 
3082   const FormatToken &InitialToken = *FormatTok;
3083   nextToken();
3084   if (InitialToken.is(TT_NamespaceMacro)) {
3085     parseParens();
3086   } else {
3087     while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::kw_inline,
3088                               tok::l_square, tok::period, tok::l_paren) ||
3089            (Style.isCSharp() && FormatTok->is(tok::kw_union))) {
3090       if (FormatTok->is(tok::l_square))
3091         parseSquare();
3092       else if (FormatTok->is(tok::l_paren))
3093         parseParens();
3094       else
3095         nextToken();
3096     }
3097   }
3098   if (FormatTok->is(tok::l_brace)) {
3099     FormatTok->setFinalizedType(TT_NamespaceLBrace);
3100 
3101     if (ShouldBreakBeforeBrace(Style, InitialToken))
3102       addUnwrappedLine();
3103 
3104     unsigned AddLevels =
3105         Style.NamespaceIndentation == FormatStyle::NI_All ||
3106                 (Style.NamespaceIndentation == FormatStyle::NI_Inner &&
3107                  DeclarationScopeStack.size() > 1)
3108             ? 1u
3109             : 0u;
3110     bool ManageWhitesmithsBraces =
3111         AddLevels == 0u &&
3112         Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths;
3113 
3114     // If we're in Whitesmiths mode, indent the brace if we're not indenting
3115     // the whole block.
3116     if (ManageWhitesmithsBraces)
3117       ++Line->Level;
3118 
3119     // Munch the semicolon after a namespace. This is more common than one would
3120     // think. Putting the semicolon into its own line is very ugly.
3121     parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/true,
3122                /*KeepBraces=*/true, /*IfKind=*/nullptr,
3123                ManageWhitesmithsBraces);
3124 
3125     addUnwrappedLine(AddLevels > 0 ? LineLevel::Remove : LineLevel::Keep);
3126 
3127     if (ManageWhitesmithsBraces)
3128       --Line->Level;
3129   }
3130   // FIXME: Add error handling.
3131 }
3132 
3133 void UnwrappedLineParser::parseNew() {
3134   assert(FormatTok->is(tok::kw_new) && "'new' expected");
3135   nextToken();
3136 
3137   if (Style.isCSharp()) {
3138     do {
3139       // Handle constructor invocation, e.g. `new(field: value)`.
3140       if (FormatTok->is(tok::l_paren))
3141         parseParens();
3142 
3143       // Handle array initialization syntax, e.g. `new[] {10, 20, 30}`.
3144       if (FormatTok->is(tok::l_brace))
3145         parseBracedList();
3146 
3147       if (FormatTok->isOneOf(tok::semi, tok::comma))
3148         return;
3149 
3150       nextToken();
3151     } while (!eof());
3152   }
3153 
3154   if (Style.Language != FormatStyle::LK_Java)
3155     return;
3156 
3157   // In Java, we can parse everything up to the parens, which aren't optional.
3158   do {
3159     // There should not be a ;, { or } before the new's open paren.
3160     if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace))
3161       return;
3162 
3163     // Consume the parens.
3164     if (FormatTok->is(tok::l_paren)) {
3165       parseParens();
3166 
3167       // If there is a class body of an anonymous class, consume that as child.
3168       if (FormatTok->is(tok::l_brace))
3169         parseChildBlock();
3170       return;
3171     }
3172     nextToken();
3173   } while (!eof());
3174 }
3175 
3176 void UnwrappedLineParser::parseLoopBody(bool KeepBraces, bool WrapRightBrace) {
3177   keepAncestorBraces();
3178 
3179   if (isBlockBegin(*FormatTok)) {
3180     FormatTok->setFinalizedType(TT_ControlStatementLBrace);
3181     FormatToken *LeftBrace = FormatTok;
3182     CompoundStatementIndenter Indenter(this, Style, Line->Level);
3183     parseBlock(/*MustBeDeclaration=*/false, /*AddLevels=*/1u,
3184                /*MunchSemi=*/true, KeepBraces);
3185     setPreviousRBraceType(TT_ControlStatementRBrace);
3186     if (!KeepBraces) {
3187       assert(!NestedTooDeep.empty());
3188       if (!NestedTooDeep.back())
3189         markOptionalBraces(LeftBrace);
3190     }
3191     if (WrapRightBrace)
3192       addUnwrappedLine();
3193   } else {
3194     parseUnbracedBody();
3195   }
3196 
3197   if (!KeepBraces)
3198     NestedTooDeep.pop_back();
3199 }
3200 
3201 void UnwrappedLineParser::parseForOrWhileLoop(bool HasParens) {
3202   assert((FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) ||
3203           (Style.isVerilog() &&
3204            FormatTok->isOneOf(Keywords.kw_always, Keywords.kw_always_comb,
3205                               Keywords.kw_always_ff, Keywords.kw_always_latch,
3206                               Keywords.kw_final, Keywords.kw_initial,
3207                               Keywords.kw_foreach, Keywords.kw_forever,
3208                               Keywords.kw_repeat))) &&
3209          "'for', 'while' or foreach macro expected");
3210   const bool KeepBraces = !Style.RemoveBracesLLVM ||
3211                           !FormatTok->isOneOf(tok::kw_for, tok::kw_while);
3212 
3213   nextToken();
3214   // JS' for await ( ...
3215   if (Style.isJavaScript() && FormatTok->is(Keywords.kw_await))
3216     nextToken();
3217   if (IsCpp && FormatTok->is(tok::kw_co_await))
3218     nextToken();
3219   if (HasParens && FormatTok->is(tok::l_paren)) {
3220     // The type is only set for Verilog basically because we were afraid to
3221     // change the existing behavior for loops. See the discussion on D121756 for
3222     // details.
3223     if (Style.isVerilog())
3224       FormatTok->setFinalizedType(TT_ConditionLParen);
3225     parseParens();
3226   }
3227 
3228   if (Style.isVerilog()) {
3229     // Event control.
3230     parseVerilogSensitivityList();
3231   } else if (Style.AllowShortLoopsOnASingleLine && FormatTok->is(tok::semi) &&
3232              Tokens->getPreviousToken()->is(tok::r_paren)) {
3233     nextToken();
3234     addUnwrappedLine();
3235     return;
3236   }
3237 
3238   handleAttributes();
3239   parseLoopBody(KeepBraces, /*WrapRightBrace=*/true);
3240 }
3241 
3242 void UnwrappedLineParser::parseDoWhile() {
3243   assert(FormatTok->is(tok::kw_do) && "'do' expected");
3244   nextToken();
3245 
3246   parseLoopBody(/*KeepBraces=*/true, Style.BraceWrapping.BeforeWhile);
3247 
3248   // FIXME: Add error handling.
3249   if (FormatTok->isNot(tok::kw_while)) {
3250     addUnwrappedLine();
3251     return;
3252   }
3253 
3254   FormatTok->setFinalizedType(TT_DoWhile);
3255 
3256   // If in Whitesmiths mode, the line with the while() needs to be indented
3257   // to the same level as the block.
3258   if (Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths)
3259     ++Line->Level;
3260 
3261   nextToken();
3262   parseStructuralElement();
3263 }
3264 
3265 void UnwrappedLineParser::parseLabel(bool LeftAlignLabel) {
3266   nextToken();
3267   unsigned OldLineLevel = Line->Level;
3268 
3269   if (LeftAlignLabel)
3270     Line->Level = 0;
3271   else if (Line->Level > 1 || (!Line->InPPDirective && Line->Level > 0))
3272     --Line->Level;
3273 
3274   if (!Style.IndentCaseBlocks && CommentsBeforeNextToken.empty() &&
3275       FormatTok->is(tok::l_brace)) {
3276 
3277     CompoundStatementIndenter Indenter(this, Line->Level,
3278                                        Style.BraceWrapping.AfterCaseLabel,
3279                                        Style.BraceWrapping.IndentBraces);
3280     parseBlock();
3281     if (FormatTok->is(tok::kw_break)) {
3282       if (Style.BraceWrapping.AfterControlStatement ==
3283           FormatStyle::BWACS_Always) {
3284         addUnwrappedLine();
3285         if (!Style.IndentCaseBlocks &&
3286             Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths) {
3287           ++Line->Level;
3288         }
3289       }
3290       parseStructuralElement();
3291     }
3292     addUnwrappedLine();
3293   } else {
3294     if (FormatTok->is(tok::semi))
3295       nextToken();
3296     addUnwrappedLine();
3297   }
3298   Line->Level = OldLineLevel;
3299   if (FormatTok->isNot(tok::l_brace)) {
3300     parseStructuralElement();
3301     addUnwrappedLine();
3302   }
3303 }
3304 
3305 void UnwrappedLineParser::parseCaseLabel() {
3306   assert(FormatTok->is(tok::kw_case) && "'case' expected");
3307   auto *Case = FormatTok;
3308 
3309   // FIXME: fix handling of complex expressions here.
3310   do {
3311     nextToken();
3312     if (FormatTok->is(tok::colon)) {
3313       FormatTok->setFinalizedType(TT_CaseLabelColon);
3314       break;
3315     }
3316     if (Style.Language == FormatStyle::LK_Java && FormatTok->is(tok::arrow)) {
3317       FormatTok->setFinalizedType(TT_CaseLabelArrow);
3318       Case->setFinalizedType(TT_SwitchExpressionLabel);
3319       break;
3320     }
3321   } while (!eof());
3322   parseLabel();
3323 }
3324 
3325 void UnwrappedLineParser::parseSwitch(bool IsExpr) {
3326   assert(FormatTok->is(tok::kw_switch) && "'switch' expected");
3327   nextToken();
3328   if (FormatTok->is(tok::l_paren))
3329     parseParens();
3330 
3331   keepAncestorBraces();
3332 
3333   if (FormatTok->is(tok::l_brace)) {
3334     CompoundStatementIndenter Indenter(this, Style, Line->Level);
3335     FormatTok->setFinalizedType(IsExpr ? TT_SwitchExpressionLBrace
3336                                        : TT_ControlStatementLBrace);
3337     if (IsExpr)
3338       parseChildBlock();
3339     else
3340       parseBlock();
3341     setPreviousRBraceType(TT_ControlStatementRBrace);
3342     if (!IsExpr)
3343       addUnwrappedLine();
3344   } else {
3345     addUnwrappedLine();
3346     ++Line->Level;
3347     parseStructuralElement();
3348     --Line->Level;
3349   }
3350 
3351   if (Style.RemoveBracesLLVM)
3352     NestedTooDeep.pop_back();
3353 }
3354 
3355 // Operators that can follow a C variable.
3356 static bool isCOperatorFollowingVar(tok::TokenKind Kind) {
3357   switch (Kind) {
3358   case tok::ampamp:
3359   case tok::ampequal:
3360   case tok::arrow:
3361   case tok::caret:
3362   case tok::caretequal:
3363   case tok::comma:
3364   case tok::ellipsis:
3365   case tok::equal:
3366   case tok::equalequal:
3367   case tok::exclaim:
3368   case tok::exclaimequal:
3369   case tok::greater:
3370   case tok::greaterequal:
3371   case tok::greatergreater:
3372   case tok::greatergreaterequal:
3373   case tok::l_paren:
3374   case tok::l_square:
3375   case tok::less:
3376   case tok::lessequal:
3377   case tok::lessless:
3378   case tok::lesslessequal:
3379   case tok::minus:
3380   case tok::minusequal:
3381   case tok::minusminus:
3382   case tok::percent:
3383   case tok::percentequal:
3384   case tok::period:
3385   case tok::pipe:
3386   case tok::pipeequal:
3387   case tok::pipepipe:
3388   case tok::plus:
3389   case tok::plusequal:
3390   case tok::plusplus:
3391   case tok::question:
3392   case tok::r_brace:
3393   case tok::r_paren:
3394   case tok::r_square:
3395   case tok::semi:
3396   case tok::slash:
3397   case tok::slashequal:
3398   case tok::star:
3399   case tok::starequal:
3400     return true;
3401   default:
3402     return false;
3403   }
3404 }
3405 
3406 void UnwrappedLineParser::parseAccessSpecifier() {
3407   FormatToken *AccessSpecifierCandidate = FormatTok;
3408   nextToken();
3409   // Understand Qt's slots.
3410   if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots))
3411     nextToken();
3412   // Otherwise, we don't know what it is, and we'd better keep the next token.
3413   if (FormatTok->is(tok::colon)) {
3414     nextToken();
3415     addUnwrappedLine();
3416   } else if (FormatTok->isNot(tok::coloncolon) &&
3417              !isCOperatorFollowingVar(FormatTok->Tok.getKind())) {
3418     // Not a variable name nor namespace name.
3419     addUnwrappedLine();
3420   } else if (AccessSpecifierCandidate) {
3421     // Consider the access specifier to be a C identifier.
3422     AccessSpecifierCandidate->Tok.setKind(tok::identifier);
3423   }
3424 }
3425 
3426 /// \brief Parses a requires, decides if it is a clause or an expression.
3427 /// \pre The current token has to be the requires keyword.
3428 /// \returns true if it parsed a clause.
3429 bool UnwrappedLineParser::parseRequires() {
3430   assert(FormatTok->is(tok::kw_requires) && "'requires' expected");
3431   auto RequiresToken = FormatTok;
3432 
3433   // We try to guess if it is a requires clause, or a requires expression. For
3434   // that we first consume the keyword and check the next token.
3435   nextToken();
3436 
3437   switch (FormatTok->Tok.getKind()) {
3438   case tok::l_brace:
3439     // This can only be an expression, never a clause.
3440     parseRequiresExpression(RequiresToken);
3441     return false;
3442   case tok::l_paren:
3443     // Clauses and expression can start with a paren, it's unclear what we have.
3444     break;
3445   default:
3446     // All other tokens can only be a clause.
3447     parseRequiresClause(RequiresToken);
3448     return true;
3449   }
3450 
3451   // Looking forward we would have to decide if there are function declaration
3452   // like arguments to the requires expression:
3453   // requires (T t) {
3454   // Or there is a constraint expression for the requires clause:
3455   // requires (C<T> && ...
3456 
3457   // But first let's look behind.
3458   auto *PreviousNonComment = RequiresToken->getPreviousNonComment();
3459 
3460   if (!PreviousNonComment ||
3461       PreviousNonComment->is(TT_RequiresExpressionLBrace)) {
3462     // If there is no token, or an expression left brace, we are a requires
3463     // clause within a requires expression.
3464     parseRequiresClause(RequiresToken);
3465     return true;
3466   }
3467 
3468   switch (PreviousNonComment->Tok.getKind()) {
3469   case tok::greater:
3470   case tok::r_paren:
3471   case tok::kw_noexcept:
3472   case tok::kw_const:
3473     // This is a requires clause.
3474     parseRequiresClause(RequiresToken);
3475     return true;
3476   case tok::amp:
3477   case tok::ampamp: {
3478     // This can be either:
3479     // if (... && requires (T t) ...)
3480     // Or
3481     // void member(...) && requires (C<T> ...
3482     // We check the one token before that for a const:
3483     // void member(...) const && requires (C<T> ...
3484     auto PrevPrev = PreviousNonComment->getPreviousNonComment();
3485     if (PrevPrev && PrevPrev->is(tok::kw_const)) {
3486       parseRequiresClause(RequiresToken);
3487       return true;
3488     }
3489     break;
3490   }
3491   default:
3492     if (PreviousNonComment->isTypeOrIdentifier(LangOpts)) {
3493       // This is a requires clause.
3494       parseRequiresClause(RequiresToken);
3495       return true;
3496     }
3497     // It's an expression.
3498     parseRequiresExpression(RequiresToken);
3499     return false;
3500   }
3501 
3502   // Now we look forward and try to check if the paren content is a parameter
3503   // list. The parameters can be cv-qualified and contain references or
3504   // pointers.
3505   // So we want basically to check for TYPE NAME, but TYPE can contain all kinds
3506   // of stuff: typename, const, *, &, &&, ::, identifiers.
3507 
3508   unsigned StoredPosition = Tokens->getPosition();
3509   FormatToken *NextToken = Tokens->getNextToken();
3510   int Lookahead = 0;
3511   auto PeekNext = [&Lookahead, &NextToken, this] {
3512     ++Lookahead;
3513     NextToken = Tokens->getNextToken();
3514   };
3515 
3516   bool FoundType = false;
3517   bool LastWasColonColon = false;
3518   int OpenAngles = 0;
3519 
3520   for (; Lookahead < 50; PeekNext()) {
3521     switch (NextToken->Tok.getKind()) {
3522     case tok::kw_volatile:
3523     case tok::kw_const:
3524     case tok::comma:
3525       if (OpenAngles == 0) {
3526         FormatTok = Tokens->setPosition(StoredPosition);
3527         parseRequiresExpression(RequiresToken);
3528         return false;
3529       }
3530       break;
3531     case tok::eof:
3532       // Break out of the loop.
3533       Lookahead = 50;
3534       break;
3535     case tok::coloncolon:
3536       LastWasColonColon = true;
3537       break;
3538     case tok::kw_decltype:
3539     case tok::identifier:
3540       if (FoundType && !LastWasColonColon && OpenAngles == 0) {
3541         FormatTok = Tokens->setPosition(StoredPosition);
3542         parseRequiresExpression(RequiresToken);
3543         return false;
3544       }
3545       FoundType = true;
3546       LastWasColonColon = false;
3547       break;
3548     case tok::less:
3549       ++OpenAngles;
3550       break;
3551     case tok::greater:
3552       --OpenAngles;
3553       break;
3554     default:
3555       if (NextToken->isTypeName(LangOpts)) {
3556         FormatTok = Tokens->setPosition(StoredPosition);
3557         parseRequiresExpression(RequiresToken);
3558         return false;
3559       }
3560       break;
3561     }
3562   }
3563   // This seems to be a complicated expression, just assume it's a clause.
3564   FormatTok = Tokens->setPosition(StoredPosition);
3565   parseRequiresClause(RequiresToken);
3566   return true;
3567 }
3568 
3569 /// \brief Parses a requires clause.
3570 /// \param RequiresToken The requires keyword token, which starts this clause.
3571 /// \pre We need to be on the next token after the requires keyword.
3572 /// \sa parseRequiresExpression
3573 ///
3574 /// Returns if it either has finished parsing the clause, or it detects, that
3575 /// the clause is incorrect.
3576 void UnwrappedLineParser::parseRequiresClause(FormatToken *RequiresToken) {
3577   assert(FormatTok->getPreviousNonComment() == RequiresToken);
3578   assert(RequiresToken->is(tok::kw_requires) && "'requires' expected");
3579 
3580   // If there is no previous token, we are within a requires expression,
3581   // otherwise we will always have the template or function declaration in front
3582   // of it.
3583   bool InRequiresExpression =
3584       !RequiresToken->Previous ||
3585       RequiresToken->Previous->is(TT_RequiresExpressionLBrace);
3586 
3587   RequiresToken->setFinalizedType(InRequiresExpression
3588                                       ? TT_RequiresClauseInARequiresExpression
3589                                       : TT_RequiresClause);
3590 
3591   // NOTE: parseConstraintExpression is only ever called from this function.
3592   // It could be inlined into here.
3593   parseConstraintExpression();
3594 
3595   if (!InRequiresExpression)
3596     FormatTok->Previous->ClosesRequiresClause = true;
3597 }
3598 
3599 /// \brief Parses a requires expression.
3600 /// \param RequiresToken The requires keyword token, which starts this clause.
3601 /// \pre We need to be on the next token after the requires keyword.
3602 /// \sa parseRequiresClause
3603 ///
3604 /// Returns if it either has finished parsing the expression, or it detects,
3605 /// that the expression is incorrect.
3606 void UnwrappedLineParser::parseRequiresExpression(FormatToken *RequiresToken) {
3607   assert(FormatTok->getPreviousNonComment() == RequiresToken);
3608   assert(RequiresToken->is(tok::kw_requires) && "'requires' expected");
3609 
3610   RequiresToken->setFinalizedType(TT_RequiresExpression);
3611 
3612   if (FormatTok->is(tok::l_paren)) {
3613     FormatTok->setFinalizedType(TT_RequiresExpressionLParen);
3614     parseParens();
3615   }
3616 
3617   if (FormatTok->is(tok::l_brace)) {
3618     FormatTok->setFinalizedType(TT_RequiresExpressionLBrace);
3619     parseChildBlock();
3620   }
3621 }
3622 
3623 /// \brief Parses a constraint expression.
3624 ///
3625 /// This is the body of a requires clause. It returns, when the parsing is
3626 /// complete, or the expression is incorrect.
3627 void UnwrappedLineParser::parseConstraintExpression() {
3628   // The special handling for lambdas is needed since tryToParseLambda() eats a
3629   // token and if a requires expression is the last part of a requires clause
3630   // and followed by an attribute like [[nodiscard]] the ClosesRequiresClause is
3631   // not set on the correct token. Thus we need to be aware if we even expect a
3632   // lambda to be possible.
3633   // template <typename T> requires requires { ... } [[nodiscard]] ...;
3634   bool LambdaNextTimeAllowed = true;
3635 
3636   // Within lambda declarations, it is permitted to put a requires clause after
3637   // its template parameter list, which would place the requires clause right
3638   // before the parentheses of the parameters of the lambda declaration. Thus,
3639   // we track if we expect to see grouping parentheses at all.
3640   // Without this check, `requires foo<T> (T t)` in the below example would be
3641   // seen as the whole requires clause, accidentally eating the parameters of
3642   // the lambda.
3643   // [&]<typename T> requires foo<T> (T t) { ... };
3644   bool TopLevelParensAllowed = true;
3645 
3646   do {
3647     bool LambdaThisTimeAllowed = std::exchange(LambdaNextTimeAllowed, false);
3648 
3649     switch (FormatTok->Tok.getKind()) {
3650     case tok::kw_requires: {
3651       auto RequiresToken = FormatTok;
3652       nextToken();
3653       parseRequiresExpression(RequiresToken);
3654       break;
3655     }
3656 
3657     case tok::l_paren:
3658       if (!TopLevelParensAllowed)
3659         return;
3660       parseParens(/*AmpAmpTokenType=*/TT_BinaryOperator);
3661       TopLevelParensAllowed = false;
3662       break;
3663 
3664     case tok::l_square:
3665       if (!LambdaThisTimeAllowed || !tryToParseLambda())
3666         return;
3667       break;
3668 
3669     case tok::kw_const:
3670     case tok::semi:
3671     case tok::kw_class:
3672     case tok::kw_struct:
3673     case tok::kw_union:
3674       return;
3675 
3676     case tok::l_brace:
3677       // Potential function body.
3678       return;
3679 
3680     case tok::ampamp:
3681     case tok::pipepipe:
3682       FormatTok->setFinalizedType(TT_BinaryOperator);
3683       nextToken();
3684       LambdaNextTimeAllowed = true;
3685       TopLevelParensAllowed = true;
3686       break;
3687 
3688     case tok::comma:
3689     case tok::comment:
3690       LambdaNextTimeAllowed = LambdaThisTimeAllowed;
3691       nextToken();
3692       break;
3693 
3694     case tok::kw_sizeof:
3695     case tok::greater:
3696     case tok::greaterequal:
3697     case tok::greatergreater:
3698     case tok::less:
3699     case tok::lessequal:
3700     case tok::lessless:
3701     case tok::equalequal:
3702     case tok::exclaim:
3703     case tok::exclaimequal:
3704     case tok::plus:
3705     case tok::minus:
3706     case tok::star:
3707     case tok::slash:
3708       LambdaNextTimeAllowed = true;
3709       TopLevelParensAllowed = true;
3710       // Just eat them.
3711       nextToken();
3712       break;
3713 
3714     case tok::numeric_constant:
3715     case tok::coloncolon:
3716     case tok::kw_true:
3717     case tok::kw_false:
3718       TopLevelParensAllowed = false;
3719       // Just eat them.
3720       nextToken();
3721       break;
3722 
3723     case tok::kw_static_cast:
3724     case tok::kw_const_cast:
3725     case tok::kw_reinterpret_cast:
3726     case tok::kw_dynamic_cast:
3727       nextToken();
3728       if (FormatTok->isNot(tok::less))
3729         return;
3730 
3731       nextToken();
3732       parseBracedList(/*IsAngleBracket=*/true);
3733       break;
3734 
3735     default:
3736       if (!FormatTok->Tok.getIdentifierInfo()) {
3737         // Identifiers are part of the default case, we check for more then
3738         // tok::identifier to handle builtin type traits.
3739         return;
3740       }
3741 
3742       // We need to differentiate identifiers for a template deduction guide,
3743       // variables, or function return types (the constraint expression has
3744       // ended before that), and basically all other cases. But it's easier to
3745       // check the other way around.
3746       assert(FormatTok->Previous);
3747       switch (FormatTok->Previous->Tok.getKind()) {
3748       case tok::coloncolon:  // Nested identifier.
3749       case tok::ampamp:      // Start of a function or variable for the
3750       case tok::pipepipe:    // constraint expression. (binary)
3751       case tok::exclaim:     // The same as above, but unary.
3752       case tok::kw_requires: // Initial identifier of a requires clause.
3753       case tok::equal:       // Initial identifier of a concept declaration.
3754         break;
3755       default:
3756         return;
3757       }
3758 
3759       // Read identifier with optional template declaration.
3760       nextToken();
3761       if (FormatTok->is(tok::less)) {
3762         nextToken();
3763         parseBracedList(/*IsAngleBracket=*/true);
3764       }
3765       TopLevelParensAllowed = false;
3766       break;
3767     }
3768   } while (!eof());
3769 }
3770 
3771 bool UnwrappedLineParser::parseEnum() {
3772   const FormatToken &InitialToken = *FormatTok;
3773 
3774   // Won't be 'enum' for NS_ENUMs.
3775   if (FormatTok->is(tok::kw_enum))
3776     nextToken();
3777 
3778   // In TypeScript, "enum" can also be used as property name, e.g. in interface
3779   // declarations. An "enum" keyword followed by a colon would be a syntax
3780   // error and thus assume it is just an identifier.
3781   if (Style.isJavaScript() && FormatTok->isOneOf(tok::colon, tok::question))
3782     return false;
3783 
3784   // In protobuf, "enum" can be used as a field name.
3785   if (Style.Language == FormatStyle::LK_Proto && FormatTok->is(tok::equal))
3786     return false;
3787 
3788   if (IsCpp) {
3789     // Eat up enum class ...
3790     if (FormatTok->isOneOf(tok::kw_class, tok::kw_struct))
3791       nextToken();
3792     while (FormatTok->is(tok::l_square))
3793       if (!handleCppAttributes())
3794         return false;
3795   }
3796 
3797   while (FormatTok->Tok.getIdentifierInfo() ||
3798          FormatTok->isOneOf(tok::colon, tok::coloncolon, tok::less,
3799                             tok::greater, tok::comma, tok::question,
3800                             tok::l_square)) {
3801     if (Style.isVerilog()) {
3802       FormatTok->setFinalizedType(TT_VerilogDimensionedTypeName);
3803       nextToken();
3804       // In Verilog the base type can have dimensions.
3805       while (FormatTok->is(tok::l_square))
3806         parseSquare();
3807     } else {
3808       nextToken();
3809     }
3810     // We can have macros or attributes in between 'enum' and the enum name.
3811     if (FormatTok->is(tok::l_paren))
3812       parseParens();
3813     if (FormatTok->is(tok::identifier)) {
3814       nextToken();
3815       // If there are two identifiers in a row, this is likely an elaborate
3816       // return type. In Java, this can be "implements", etc.
3817       if (IsCpp && FormatTok->is(tok::identifier))
3818         return false;
3819     }
3820   }
3821 
3822   // Just a declaration or something is wrong.
3823   if (FormatTok->isNot(tok::l_brace))
3824     return true;
3825   FormatTok->setFinalizedType(TT_EnumLBrace);
3826   FormatTok->setBlockKind(BK_Block);
3827 
3828   if (Style.Language == FormatStyle::LK_Java) {
3829     // Java enums are different.
3830     parseJavaEnumBody();
3831     return true;
3832   }
3833   if (Style.Language == FormatStyle::LK_Proto) {
3834     parseBlock(/*MustBeDeclaration=*/true);
3835     return true;
3836   }
3837 
3838   if (!Style.AllowShortEnumsOnASingleLine &&
3839       ShouldBreakBeforeBrace(Style, InitialToken)) {
3840     addUnwrappedLine();
3841   }
3842   // Parse enum body.
3843   nextToken();
3844   if (!Style.AllowShortEnumsOnASingleLine) {
3845     addUnwrappedLine();
3846     Line->Level += 1;
3847   }
3848   bool HasError = !parseBracedList(/*IsAngleBracket=*/false, /*IsEnum=*/true);
3849   if (!Style.AllowShortEnumsOnASingleLine)
3850     Line->Level -= 1;
3851   if (HasError) {
3852     if (FormatTok->is(tok::semi))
3853       nextToken();
3854     addUnwrappedLine();
3855   }
3856   setPreviousRBraceType(TT_EnumRBrace);
3857   return true;
3858 
3859   // There is no addUnwrappedLine() here so that we fall through to parsing a
3860   // structural element afterwards. Thus, in "enum A {} n, m;",
3861   // "} n, m;" will end up in one unwrapped line.
3862 }
3863 
3864 bool UnwrappedLineParser::parseStructLike() {
3865   // parseRecord falls through and does not yet add an unwrapped line as a
3866   // record declaration or definition can start a structural element.
3867   parseRecord();
3868   // This does not apply to Java, JavaScript and C#.
3869   if (Style.Language == FormatStyle::LK_Java || Style.isJavaScript() ||
3870       Style.isCSharp()) {
3871     if (FormatTok->is(tok::semi))
3872       nextToken();
3873     addUnwrappedLine();
3874     return true;
3875   }
3876   return false;
3877 }
3878 
3879 namespace {
3880 // A class used to set and restore the Token position when peeking
3881 // ahead in the token source.
3882 class ScopedTokenPosition {
3883   unsigned StoredPosition;
3884   FormatTokenSource *Tokens;
3885 
3886 public:
3887   ScopedTokenPosition(FormatTokenSource *Tokens) : Tokens(Tokens) {
3888     assert(Tokens && "Tokens expected to not be null");
3889     StoredPosition = Tokens->getPosition();
3890   }
3891 
3892   ~ScopedTokenPosition() { Tokens->setPosition(StoredPosition); }
3893 };
3894 } // namespace
3895 
3896 // Look to see if we have [[ by looking ahead, if
3897 // its not then rewind to the original position.
3898 bool UnwrappedLineParser::tryToParseSimpleAttribute() {
3899   ScopedTokenPosition AutoPosition(Tokens);
3900   FormatToken *Tok = Tokens->getNextToken();
3901   // We already read the first [ check for the second.
3902   if (Tok->isNot(tok::l_square))
3903     return false;
3904   // Double check that the attribute is just something
3905   // fairly simple.
3906   while (Tok->isNot(tok::eof)) {
3907     if (Tok->is(tok::r_square))
3908       break;
3909     Tok = Tokens->getNextToken();
3910   }
3911   if (Tok->is(tok::eof))
3912     return false;
3913   Tok = Tokens->getNextToken();
3914   if (Tok->isNot(tok::r_square))
3915     return false;
3916   Tok = Tokens->getNextToken();
3917   if (Tok->is(tok::semi))
3918     return false;
3919   return true;
3920 }
3921 
3922 void UnwrappedLineParser::parseJavaEnumBody() {
3923   assert(FormatTok->is(tok::l_brace));
3924   const FormatToken *OpeningBrace = FormatTok;
3925 
3926   // Determine whether the enum is simple, i.e. does not have a semicolon or
3927   // constants with class bodies. Simple enums can be formatted like braced
3928   // lists, contracted to a single line, etc.
3929   unsigned StoredPosition = Tokens->getPosition();
3930   bool IsSimple = true;
3931   FormatToken *Tok = Tokens->getNextToken();
3932   while (Tok->isNot(tok::eof)) {
3933     if (Tok->is(tok::r_brace))
3934       break;
3935     if (Tok->isOneOf(tok::l_brace, tok::semi)) {
3936       IsSimple = false;
3937       break;
3938     }
3939     // FIXME: This will also mark enums with braces in the arguments to enum
3940     // constants as "not simple". This is probably fine in practice, though.
3941     Tok = Tokens->getNextToken();
3942   }
3943   FormatTok = Tokens->setPosition(StoredPosition);
3944 
3945   if (IsSimple) {
3946     nextToken();
3947     parseBracedList();
3948     addUnwrappedLine();
3949     return;
3950   }
3951 
3952   // Parse the body of a more complex enum.
3953   // First add a line for everything up to the "{".
3954   nextToken();
3955   addUnwrappedLine();
3956   ++Line->Level;
3957 
3958   // Parse the enum constants.
3959   while (!eof()) {
3960     if (FormatTok->is(tok::l_brace)) {
3961       // Parse the constant's class body.
3962       parseBlock(/*MustBeDeclaration=*/true, /*AddLevels=*/1u,
3963                  /*MunchSemi=*/false);
3964     } else if (FormatTok->is(tok::l_paren)) {
3965       parseParens();
3966     } else if (FormatTok->is(tok::comma)) {
3967       nextToken();
3968       addUnwrappedLine();
3969     } else if (FormatTok->is(tok::semi)) {
3970       nextToken();
3971       addUnwrappedLine();
3972       break;
3973     } else if (FormatTok->is(tok::r_brace)) {
3974       addUnwrappedLine();
3975       break;
3976     } else {
3977       nextToken();
3978     }
3979   }
3980 
3981   // Parse the class body after the enum's ";" if any.
3982   parseLevel(OpeningBrace);
3983   nextToken();
3984   --Line->Level;
3985   addUnwrappedLine();
3986 }
3987 
3988 void UnwrappedLineParser::parseRecord(bool ParseAsExpr) {
3989   const FormatToken &InitialToken = *FormatTok;
3990   nextToken();
3991 
3992   const FormatToken *ClassName = nullptr;
3993   bool IsDerived = false;
3994   auto IsNonMacroIdentifier = [](const FormatToken *Tok) {
3995     return Tok->is(tok::identifier) && Tok->TokenText != Tok->TokenText.upper();
3996   };
3997   // JavaScript/TypeScript supports anonymous classes like:
3998   // a = class extends foo { }
3999   bool JSPastExtendsOrImplements = false;
4000   // The actual identifier can be a nested name specifier, and in macros
4001   // it is often token-pasted.
4002   // An [[attribute]] can be before the identifier.
4003   while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash,
4004                             tok::kw_alignas, tok::l_square) ||
4005          FormatTok->isAttribute() ||
4006          ((Style.Language == FormatStyle::LK_Java || Style.isJavaScript()) &&
4007           FormatTok->isOneOf(tok::period, tok::comma))) {
4008     if (Style.isJavaScript() &&
4009         FormatTok->isOneOf(Keywords.kw_extends, Keywords.kw_implements)) {
4010       JSPastExtendsOrImplements = true;
4011       // JavaScript/TypeScript supports inline object types in
4012       // extends/implements positions:
4013       //     class Foo implements {bar: number} { }
4014       nextToken();
4015       if (FormatTok->is(tok::l_brace)) {
4016         tryToParseBracedList();
4017         continue;
4018       }
4019     }
4020     if (FormatTok->is(tok::l_square) && handleCppAttributes())
4021       continue;
4022     const auto *Previous = FormatTok;
4023     nextToken();
4024     switch (FormatTok->Tok.getKind()) {
4025     case tok::l_paren:
4026       // We can have macros in between 'class' and the class name.
4027       if (!IsNonMacroIdentifier(Previous) ||
4028           // e.g. `struct macro(a) S { int i; };`
4029           Previous->Previous == &InitialToken) {
4030         parseParens();
4031       }
4032       break;
4033     case tok::coloncolon:
4034     case tok::hashhash:
4035       break;
4036     default:
4037       if (!JSPastExtendsOrImplements && !ClassName &&
4038           Previous->is(tok::identifier) && Previous->isNot(TT_AttributeMacro)) {
4039         ClassName = Previous;
4040       }
4041     }
4042   }
4043 
4044   auto IsListInitialization = [&] {
4045     if (!ClassName || IsDerived)
4046       return false;
4047     assert(FormatTok->is(tok::l_brace));
4048     const auto *Prev = FormatTok->getPreviousNonComment();
4049     assert(Prev);
4050     return Prev != ClassName && Prev->is(tok::identifier) &&
4051            Prev->isNot(Keywords.kw_final) && tryToParseBracedList();
4052   };
4053 
4054   if (FormatTok->isOneOf(tok::colon, tok::less)) {
4055     int AngleNestingLevel = 0;
4056     do {
4057       if (FormatTok->is(tok::less))
4058         ++AngleNestingLevel;
4059       else if (FormatTok->is(tok::greater))
4060         --AngleNestingLevel;
4061 
4062       if (AngleNestingLevel == 0) {
4063         if (FormatTok->is(tok::colon)) {
4064           IsDerived = true;
4065         } else if (FormatTok->is(tok::identifier) &&
4066                    FormatTok->Previous->is(tok::coloncolon)) {
4067           ClassName = FormatTok;
4068         } else if (FormatTok->is(tok::l_paren) &&
4069                    IsNonMacroIdentifier(FormatTok->Previous)) {
4070           break;
4071         }
4072       }
4073       if (FormatTok->is(tok::l_brace)) {
4074         if (AngleNestingLevel == 0 && IsListInitialization())
4075           return;
4076         calculateBraceTypes(/*ExpectClassBody=*/true);
4077         if (!tryToParseBracedList())
4078           break;
4079       }
4080       if (FormatTok->is(tok::l_square)) {
4081         FormatToken *Previous = FormatTok->Previous;
4082         if (!Previous || (Previous->isNot(tok::r_paren) &&
4083                           !Previous->isTypeOrIdentifier(LangOpts))) {
4084           // Don't try parsing a lambda if we had a closing parenthesis before,
4085           // it was probably a pointer to an array: int (*)[].
4086           if (!tryToParseLambda())
4087             continue;
4088         } else {
4089           parseSquare();
4090           continue;
4091         }
4092       }
4093       if (FormatTok->is(tok::semi))
4094         return;
4095       if (Style.isCSharp() && FormatTok->is(Keywords.kw_where)) {
4096         addUnwrappedLine();
4097         nextToken();
4098         parseCSharpGenericTypeConstraint();
4099         break;
4100       }
4101       nextToken();
4102     } while (!eof());
4103   }
4104 
4105   auto GetBraceTypes =
4106       [](const FormatToken &RecordTok) -> std::pair<TokenType, TokenType> {
4107     switch (RecordTok.Tok.getKind()) {
4108     case tok::kw_class:
4109       return {TT_ClassLBrace, TT_ClassRBrace};
4110     case tok::kw_struct:
4111       return {TT_StructLBrace, TT_StructRBrace};
4112     case tok::kw_union:
4113       return {TT_UnionLBrace, TT_UnionRBrace};
4114     default:
4115       // Useful for e.g. interface.
4116       return {TT_RecordLBrace, TT_RecordRBrace};
4117     }
4118   };
4119   if (FormatTok->is(tok::l_brace)) {
4120     if (IsListInitialization())
4121       return;
4122     auto [OpenBraceType, ClosingBraceType] = GetBraceTypes(InitialToken);
4123     FormatTok->setFinalizedType(OpenBraceType);
4124     if (ParseAsExpr) {
4125       parseChildBlock();
4126     } else {
4127       if (ShouldBreakBeforeBrace(Style, InitialToken))
4128         addUnwrappedLine();
4129 
4130       unsigned AddLevels = Style.IndentAccessModifiers ? 2u : 1u;
4131       parseBlock(/*MustBeDeclaration=*/true, AddLevels, /*MunchSemi=*/false);
4132     }
4133     setPreviousRBraceType(ClosingBraceType);
4134   }
4135   // There is no addUnwrappedLine() here so that we fall through to parsing a
4136   // structural element afterwards. Thus, in "class A {} n, m;",
4137   // "} n, m;" will end up in one unwrapped line.
4138 }
4139 
4140 void UnwrappedLineParser::parseObjCMethod() {
4141   assert(FormatTok->isOneOf(tok::l_paren, tok::identifier) &&
4142          "'(' or identifier expected.");
4143   do {
4144     if (FormatTok->is(tok::semi)) {
4145       nextToken();
4146       addUnwrappedLine();
4147       return;
4148     } else if (FormatTok->is(tok::l_brace)) {
4149       if (Style.BraceWrapping.AfterFunction)
4150         addUnwrappedLine();
4151       parseBlock();
4152       addUnwrappedLine();
4153       return;
4154     } else {
4155       nextToken();
4156     }
4157   } while (!eof());
4158 }
4159 
4160 void UnwrappedLineParser::parseObjCProtocolList() {
4161   assert(FormatTok->is(tok::less) && "'<' expected.");
4162   do {
4163     nextToken();
4164     // Early exit in case someone forgot a close angle.
4165     if (FormatTok->isOneOf(tok::semi, tok::l_brace) ||
4166         FormatTok->isObjCAtKeyword(tok::objc_end)) {
4167       return;
4168     }
4169   } while (!eof() && FormatTok->isNot(tok::greater));
4170   nextToken(); // Skip '>'.
4171 }
4172 
4173 void UnwrappedLineParser::parseObjCUntilAtEnd() {
4174   do {
4175     if (FormatTok->isObjCAtKeyword(tok::objc_end)) {
4176       nextToken();
4177       addUnwrappedLine();
4178       break;
4179     }
4180     if (FormatTok->is(tok::l_brace)) {
4181       parseBlock();
4182       // In ObjC interfaces, nothing should be following the "}".
4183       addUnwrappedLine();
4184     } else if (FormatTok->is(tok::r_brace)) {
4185       // Ignore stray "}". parseStructuralElement doesn't consume them.
4186       nextToken();
4187       addUnwrappedLine();
4188     } else if (FormatTok->isOneOf(tok::minus, tok::plus)) {
4189       nextToken();
4190       parseObjCMethod();
4191     } else {
4192       parseStructuralElement();
4193     }
4194   } while (!eof());
4195 }
4196 
4197 void UnwrappedLineParser::parseObjCInterfaceOrImplementation() {
4198   assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_interface ||
4199          FormatTok->Tok.getObjCKeywordID() == tok::objc_implementation);
4200   nextToken();
4201   nextToken(); // interface name
4202 
4203   // @interface can be followed by a lightweight generic
4204   // specialization list, then either a base class or a category.
4205   if (FormatTok->is(tok::less))
4206     parseObjCLightweightGenerics();
4207   if (FormatTok->is(tok::colon)) {
4208     nextToken();
4209     nextToken(); // base class name
4210     // The base class can also have lightweight generics applied to it.
4211     if (FormatTok->is(tok::less))
4212       parseObjCLightweightGenerics();
4213   } else if (FormatTok->is(tok::l_paren)) {
4214     // Skip category, if present.
4215     parseParens();
4216   }
4217 
4218   if (FormatTok->is(tok::less))
4219     parseObjCProtocolList();
4220 
4221   if (FormatTok->is(tok::l_brace)) {
4222     if (Style.BraceWrapping.AfterObjCDeclaration)
4223       addUnwrappedLine();
4224     parseBlock(/*MustBeDeclaration=*/true);
4225   }
4226 
4227   // With instance variables, this puts '}' on its own line.  Without instance
4228   // variables, this ends the @interface line.
4229   addUnwrappedLine();
4230 
4231   parseObjCUntilAtEnd();
4232 }
4233 
4234 void UnwrappedLineParser::parseObjCLightweightGenerics() {
4235   assert(FormatTok->is(tok::less));
4236   // Unlike protocol lists, generic parameterizations support
4237   // nested angles:
4238   //
4239   // @interface Foo<ValueType : id <NSCopying, NSSecureCoding>> :
4240   //     NSObject <NSCopying, NSSecureCoding>
4241   //
4242   // so we need to count how many open angles we have left.
4243   unsigned NumOpenAngles = 1;
4244   do {
4245     nextToken();
4246     // Early exit in case someone forgot a close angle.
4247     if (FormatTok->isOneOf(tok::semi, tok::l_brace) ||
4248         FormatTok->isObjCAtKeyword(tok::objc_end)) {
4249       break;
4250     }
4251     if (FormatTok->is(tok::less)) {
4252       ++NumOpenAngles;
4253     } else if (FormatTok->is(tok::greater)) {
4254       assert(NumOpenAngles > 0 && "'>' makes NumOpenAngles negative");
4255       --NumOpenAngles;
4256     }
4257   } while (!eof() && NumOpenAngles != 0);
4258   nextToken(); // Skip '>'.
4259 }
4260 
4261 // Returns true for the declaration/definition form of @protocol,
4262 // false for the expression form.
4263 bool UnwrappedLineParser::parseObjCProtocol() {
4264   assert(FormatTok->Tok.getObjCKeywordID() == tok::objc_protocol);
4265   nextToken();
4266 
4267   if (FormatTok->is(tok::l_paren)) {
4268     // The expression form of @protocol, e.g. "Protocol* p = @protocol(foo);".
4269     return false;
4270   }
4271 
4272   // The definition/declaration form,
4273   // @protocol Foo
4274   // - (int)someMethod;
4275   // @end
4276 
4277   nextToken(); // protocol name
4278 
4279   if (FormatTok->is(tok::less))
4280     parseObjCProtocolList();
4281 
4282   // Check for protocol declaration.
4283   if (FormatTok->is(tok::semi)) {
4284     nextToken();
4285     addUnwrappedLine();
4286     return true;
4287   }
4288 
4289   addUnwrappedLine();
4290   parseObjCUntilAtEnd();
4291   return true;
4292 }
4293 
4294 void UnwrappedLineParser::parseJavaScriptEs6ImportExport() {
4295   bool IsImport = FormatTok->is(Keywords.kw_import);
4296   assert(IsImport || FormatTok->is(tok::kw_export));
4297   nextToken();
4298 
4299   // Consume the "default" in "export default class/function".
4300   if (FormatTok->is(tok::kw_default))
4301     nextToken();
4302 
4303   // Consume "async function", "function" and "default function", so that these
4304   // get parsed as free-standing JS functions, i.e. do not require a trailing
4305   // semicolon.
4306   if (FormatTok->is(Keywords.kw_async))
4307     nextToken();
4308   if (FormatTok->is(Keywords.kw_function)) {
4309     nextToken();
4310     return;
4311   }
4312 
4313   // For imports, `export *`, `export {...}`, consume the rest of the line up
4314   // to the terminating `;`. For everything else, just return and continue
4315   // parsing the structural element, i.e. the declaration or expression for
4316   // `export default`.
4317   if (!IsImport && !FormatTok->isOneOf(tok::l_brace, tok::star) &&
4318       !FormatTok->isStringLiteral() &&
4319       !(FormatTok->is(Keywords.kw_type) &&
4320         Tokens->peekNextToken()->isOneOf(tok::l_brace, tok::star))) {
4321     return;
4322   }
4323 
4324   while (!eof()) {
4325     if (FormatTok->is(tok::semi))
4326       return;
4327     if (Line->Tokens.empty()) {
4328       // Common issue: Automatic Semicolon Insertion wrapped the line, so the
4329       // import statement should terminate.
4330       return;
4331     }
4332     if (FormatTok->is(tok::l_brace)) {
4333       FormatTok->setBlockKind(BK_Block);
4334       nextToken();
4335       parseBracedList();
4336     } else {
4337       nextToken();
4338     }
4339   }
4340 }
4341 
4342 void UnwrappedLineParser::parseStatementMacro() {
4343   nextToken();
4344   if (FormatTok->is(tok::l_paren))
4345     parseParens();
4346   if (FormatTok->is(tok::semi))
4347     nextToken();
4348   addUnwrappedLine();
4349 }
4350 
4351 void UnwrappedLineParser::parseVerilogHierarchyIdentifier() {
4352   // consume things like a::`b.c[d:e] or a::*
4353   while (true) {
4354     if (FormatTok->isOneOf(tok::star, tok::period, tok::periodstar,
4355                            tok::coloncolon, tok::hash) ||
4356         Keywords.isVerilogIdentifier(*FormatTok)) {
4357       nextToken();
4358     } else if (FormatTok->is(tok::l_square)) {
4359       parseSquare();
4360     } else {
4361       break;
4362     }
4363   }
4364 }
4365 
4366 void UnwrappedLineParser::parseVerilogSensitivityList() {
4367   if (FormatTok->isNot(tok::at))
4368     return;
4369   nextToken();
4370   // A block event expression has 2 at signs.
4371   if (FormatTok->is(tok::at))
4372     nextToken();
4373   switch (FormatTok->Tok.getKind()) {
4374   case tok::star:
4375     nextToken();
4376     break;
4377   case tok::l_paren:
4378     parseParens();
4379     break;
4380   default:
4381     parseVerilogHierarchyIdentifier();
4382     break;
4383   }
4384 }
4385 
4386 unsigned UnwrappedLineParser::parseVerilogHierarchyHeader() {
4387   unsigned AddLevels = 0;
4388 
4389   if (FormatTok->is(Keywords.kw_clocking)) {
4390     nextToken();
4391     if (Keywords.isVerilogIdentifier(*FormatTok))
4392       nextToken();
4393     parseVerilogSensitivityList();
4394     if (FormatTok->is(tok::semi))
4395       nextToken();
4396   } else if (FormatTok->isOneOf(tok::kw_case, Keywords.kw_casex,
4397                                 Keywords.kw_casez, Keywords.kw_randcase,
4398                                 Keywords.kw_randsequence)) {
4399     if (Style.IndentCaseLabels)
4400       AddLevels++;
4401     nextToken();
4402     if (FormatTok->is(tok::l_paren)) {
4403       FormatTok->setFinalizedType(TT_ConditionLParen);
4404       parseParens();
4405     }
4406     if (FormatTok->isOneOf(Keywords.kw_inside, Keywords.kw_matches))
4407       nextToken();
4408     // The case header has no semicolon.
4409   } else {
4410     // "module" etc.
4411     nextToken();
4412     // all the words like the name of the module and specifiers like
4413     // "automatic" and the width of function return type
4414     while (true) {
4415       if (FormatTok->is(tok::l_square)) {
4416         auto Prev = FormatTok->getPreviousNonComment();
4417         if (Prev && Keywords.isVerilogIdentifier(*Prev))
4418           Prev->setFinalizedType(TT_VerilogDimensionedTypeName);
4419         parseSquare();
4420       } else if (Keywords.isVerilogIdentifier(*FormatTok) ||
4421                  FormatTok->isOneOf(Keywords.kw_automatic, tok::kw_static)) {
4422         nextToken();
4423       } else {
4424         break;
4425       }
4426     }
4427 
4428     auto NewLine = [this]() {
4429       addUnwrappedLine();
4430       Line->IsContinuation = true;
4431     };
4432 
4433     // package imports
4434     while (FormatTok->is(Keywords.kw_import)) {
4435       NewLine();
4436       nextToken();
4437       parseVerilogHierarchyIdentifier();
4438       if (FormatTok->is(tok::semi))
4439         nextToken();
4440     }
4441 
4442     // parameters and ports
4443     if (FormatTok->is(Keywords.kw_verilogHash)) {
4444       NewLine();
4445       nextToken();
4446       if (FormatTok->is(tok::l_paren)) {
4447         FormatTok->setFinalizedType(TT_VerilogMultiLineListLParen);
4448         parseParens();
4449       }
4450     }
4451     if (FormatTok->is(tok::l_paren)) {
4452       NewLine();
4453       FormatTok->setFinalizedType(TT_VerilogMultiLineListLParen);
4454       parseParens();
4455     }
4456 
4457     // extends and implements
4458     if (FormatTok->is(Keywords.kw_extends)) {
4459       NewLine();
4460       nextToken();
4461       parseVerilogHierarchyIdentifier();
4462       if (FormatTok->is(tok::l_paren))
4463         parseParens();
4464     }
4465     if (FormatTok->is(Keywords.kw_implements)) {
4466       NewLine();
4467       do {
4468         nextToken();
4469         parseVerilogHierarchyIdentifier();
4470       } while (FormatTok->is(tok::comma));
4471     }
4472 
4473     // Coverage event for cover groups.
4474     if (FormatTok->is(tok::at)) {
4475       NewLine();
4476       parseVerilogSensitivityList();
4477     }
4478 
4479     if (FormatTok->is(tok::semi))
4480       nextToken(/*LevelDifference=*/1);
4481     addUnwrappedLine();
4482   }
4483 
4484   return AddLevels;
4485 }
4486 
4487 void UnwrappedLineParser::parseVerilogTable() {
4488   assert(FormatTok->is(Keywords.kw_table));
4489   nextToken(/*LevelDifference=*/1);
4490   addUnwrappedLine();
4491 
4492   auto InitialLevel = Line->Level++;
4493   while (!eof() && !Keywords.isVerilogEnd(*FormatTok)) {
4494     FormatToken *Tok = FormatTok;
4495     nextToken();
4496     if (Tok->is(tok::semi))
4497       addUnwrappedLine();
4498     else if (Tok->isOneOf(tok::star, tok::colon, tok::question, tok::minus))
4499       Tok->setFinalizedType(TT_VerilogTableItem);
4500   }
4501   Line->Level = InitialLevel;
4502   nextToken(/*LevelDifference=*/-1);
4503   addUnwrappedLine();
4504 }
4505 
4506 void UnwrappedLineParser::parseVerilogCaseLabel() {
4507   // The label will get unindented in AnnotatingParser. If there are no leading
4508   // spaces, indent the rest here so that things inside the block will be
4509   // indented relative to things outside. We don't use parseLabel because we
4510   // don't know whether this colon is a label or a ternary expression at this
4511   // point.
4512   auto OrigLevel = Line->Level;
4513   auto FirstLine = CurrentLines->size();
4514   if (Line->Level == 0 || (Line->InPPDirective && Line->Level <= 1))
4515     ++Line->Level;
4516   else if (!Style.IndentCaseBlocks && Keywords.isVerilogBegin(*FormatTok))
4517     --Line->Level;
4518   parseStructuralElement();
4519   // Restore the indentation in both the new line and the line that has the
4520   // label.
4521   if (CurrentLines->size() > FirstLine)
4522     (*CurrentLines)[FirstLine].Level = OrigLevel;
4523   Line->Level = OrigLevel;
4524 }
4525 
4526 bool UnwrappedLineParser::containsExpansion(const UnwrappedLine &Line) const {
4527   for (const auto &N : Line.Tokens) {
4528     if (N.Tok->MacroCtx)
4529       return true;
4530     for (const UnwrappedLine &Child : N.Children)
4531       if (containsExpansion(Child))
4532         return true;
4533   }
4534   return false;
4535 }
4536 
4537 void UnwrappedLineParser::addUnwrappedLine(LineLevel AdjustLevel) {
4538   if (Line->Tokens.empty())
4539     return;
4540   LLVM_DEBUG({
4541     if (!parsingPPDirective()) {
4542       llvm::dbgs() << "Adding unwrapped line:\n";
4543       printDebugInfo(*Line);
4544     }
4545   });
4546 
4547   // If this line closes a block when in Whitesmiths mode, remember that
4548   // information so that the level can be decreased after the line is added.
4549   // This has to happen after the addition of the line since the line itself
4550   // needs to be indented.
4551   bool ClosesWhitesmithsBlock =
4552       Line->MatchingOpeningBlockLineIndex != UnwrappedLine::kInvalidIndex &&
4553       Style.BreakBeforeBraces == FormatStyle::BS_Whitesmiths;
4554 
4555   // If the current line was expanded from a macro call, we use it to
4556   // reconstruct an unwrapped line from the structure of the expanded unwrapped
4557   // line and the unexpanded token stream.
4558   if (!parsingPPDirective() && !InExpansion && containsExpansion(*Line)) {
4559     if (!Reconstruct)
4560       Reconstruct.emplace(Line->Level, Unexpanded);
4561     Reconstruct->addLine(*Line);
4562 
4563     // While the reconstructed unexpanded lines are stored in the normal
4564     // flow of lines, the expanded lines are stored on the side to be analyzed
4565     // in an extra step.
4566     CurrentExpandedLines.push_back(std::move(*Line));
4567 
4568     if (Reconstruct->finished()) {
4569       UnwrappedLine Reconstructed = std::move(*Reconstruct).takeResult();
4570       assert(!Reconstructed.Tokens.empty() &&
4571              "Reconstructed must at least contain the macro identifier.");
4572       assert(!parsingPPDirective());
4573       LLVM_DEBUG({
4574         llvm::dbgs() << "Adding unexpanded line:\n";
4575         printDebugInfo(Reconstructed);
4576       });
4577       ExpandedLines[Reconstructed.Tokens.begin()->Tok] = CurrentExpandedLines;
4578       Lines.push_back(std::move(Reconstructed));
4579       CurrentExpandedLines.clear();
4580       Reconstruct.reset();
4581     }
4582   } else {
4583     // At the top level we only get here when no unexpansion is going on, or
4584     // when conditional formatting led to unfinished macro reconstructions.
4585     assert(!Reconstruct || (CurrentLines != &Lines) || PPStack.size() > 0);
4586     CurrentLines->push_back(std::move(*Line));
4587   }
4588   Line->Tokens.clear();
4589   Line->MatchingOpeningBlockLineIndex = UnwrappedLine::kInvalidIndex;
4590   Line->FirstStartColumn = 0;
4591   Line->IsContinuation = false;
4592   Line->SeenDecltypeAuto = false;
4593 
4594   if (ClosesWhitesmithsBlock && AdjustLevel == LineLevel::Remove)
4595     --Line->Level;
4596   if (!parsingPPDirective() && !PreprocessorDirectives.empty()) {
4597     CurrentLines->append(
4598         std::make_move_iterator(PreprocessorDirectives.begin()),
4599         std::make_move_iterator(PreprocessorDirectives.end()));
4600     PreprocessorDirectives.clear();
4601   }
4602   // Disconnect the current token from the last token on the previous line.
4603   FormatTok->Previous = nullptr;
4604 }
4605 
4606 bool UnwrappedLineParser::eof() const { return FormatTok->is(tok::eof); }
4607 
4608 bool UnwrappedLineParser::isOnNewLine(const FormatToken &FormatTok) {
4609   return (Line->InPPDirective || FormatTok.HasUnescapedNewline) &&
4610          FormatTok.NewlinesBefore > 0;
4611 }
4612 
4613 // Checks if \p FormatTok is a line comment that continues the line comment
4614 // section on \p Line.
4615 static bool
4616 continuesLineCommentSection(const FormatToken &FormatTok,
4617                             const UnwrappedLine &Line,
4618                             const llvm::Regex &CommentPragmasRegex) {
4619   if (Line.Tokens.empty())
4620     return false;
4621 
4622   StringRef IndentContent = FormatTok.TokenText;
4623   if (FormatTok.TokenText.starts_with("//") ||
4624       FormatTok.TokenText.starts_with("/*")) {
4625     IndentContent = FormatTok.TokenText.substr(2);
4626   }
4627   if (CommentPragmasRegex.match(IndentContent))
4628     return false;
4629 
4630   // If Line starts with a line comment, then FormatTok continues the comment
4631   // section if its original column is greater or equal to the original start
4632   // column of the line.
4633   //
4634   // Define the min column token of a line as follows: if a line ends in '{' or
4635   // contains a '{' followed by a line comment, then the min column token is
4636   // that '{'. Otherwise, the min column token of the line is the first token of
4637   // the line.
4638   //
4639   // If Line starts with a token other than a line comment, then FormatTok
4640   // continues the comment section if its original column is greater than the
4641   // original start column of the min column token of the line.
4642   //
4643   // For example, the second line comment continues the first in these cases:
4644   //
4645   // // first line
4646   // // second line
4647   //
4648   // and:
4649   //
4650   // // first line
4651   //  // second line
4652   //
4653   // and:
4654   //
4655   // int i; // first line
4656   //  // second line
4657   //
4658   // and:
4659   //
4660   // do { // first line
4661   //      // second line
4662   //   int i;
4663   // } while (true);
4664   //
4665   // and:
4666   //
4667   // enum {
4668   //   a, // first line
4669   //    // second line
4670   //   b
4671   // };
4672   //
4673   // The second line comment doesn't continue the first in these cases:
4674   //
4675   //   // first line
4676   //  // second line
4677   //
4678   // and:
4679   //
4680   // int i; // first line
4681   // // second line
4682   //
4683   // and:
4684   //
4685   // do { // first line
4686   //   // second line
4687   //   int i;
4688   // } while (true);
4689   //
4690   // and:
4691   //
4692   // enum {
4693   //   a, // first line
4694   //   // second line
4695   // };
4696   const FormatToken *MinColumnToken = Line.Tokens.front().Tok;
4697 
4698   // Scan for '{//'. If found, use the column of '{' as a min column for line
4699   // comment section continuation.
4700   const FormatToken *PreviousToken = nullptr;
4701   for (const UnwrappedLineNode &Node : Line.Tokens) {
4702     if (PreviousToken && PreviousToken->is(tok::l_brace) &&
4703         isLineComment(*Node.Tok)) {
4704       MinColumnToken = PreviousToken;
4705       break;
4706     }
4707     PreviousToken = Node.Tok;
4708 
4709     // Grab the last newline preceding a token in this unwrapped line.
4710     if (Node.Tok->NewlinesBefore > 0)
4711       MinColumnToken = Node.Tok;
4712   }
4713   if (PreviousToken && PreviousToken->is(tok::l_brace))
4714     MinColumnToken = PreviousToken;
4715 
4716   return continuesLineComment(FormatTok, /*Previous=*/Line.Tokens.back().Tok,
4717                               MinColumnToken);
4718 }
4719 
4720 void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) {
4721   bool JustComments = Line->Tokens.empty();
4722   for (FormatToken *Tok : CommentsBeforeNextToken) {
4723     // Line comments that belong to the same line comment section are put on the
4724     // same line since later we might want to reflow content between them.
4725     // Additional fine-grained breaking of line comment sections is controlled
4726     // by the class BreakableLineCommentSection in case it is desirable to keep
4727     // several line comment sections in the same unwrapped line.
4728     //
4729     // FIXME: Consider putting separate line comment sections as children to the
4730     // unwrapped line instead.
4731     Tok->ContinuesLineCommentSection =
4732         continuesLineCommentSection(*Tok, *Line, CommentPragmasRegex);
4733     if (isOnNewLine(*Tok) && JustComments && !Tok->ContinuesLineCommentSection)
4734       addUnwrappedLine();
4735     pushToken(Tok);
4736   }
4737   if (NewlineBeforeNext && JustComments)
4738     addUnwrappedLine();
4739   CommentsBeforeNextToken.clear();
4740 }
4741 
4742 void UnwrappedLineParser::nextToken(int LevelDifference) {
4743   if (eof())
4744     return;
4745   flushComments(isOnNewLine(*FormatTok));
4746   pushToken(FormatTok);
4747   FormatToken *Previous = FormatTok;
4748   if (!Style.isJavaScript())
4749     readToken(LevelDifference);
4750   else
4751     readTokenWithJavaScriptASI();
4752   FormatTok->Previous = Previous;
4753   if (Style.isVerilog()) {
4754     // Blocks in Verilog can have `begin` and `end` instead of braces.  For
4755     // keywords like `begin`, we can't treat them the same as left braces
4756     // because some contexts require one of them.  For example structs use
4757     // braces and if blocks use keywords, and a left brace can occur in an if
4758     // statement, but it is not a block.  For keywords like `end`, we simply
4759     // treat them the same as right braces.
4760     if (Keywords.isVerilogEnd(*FormatTok))
4761       FormatTok->Tok.setKind(tok::r_brace);
4762   }
4763 }
4764 
4765 void UnwrappedLineParser::distributeComments(
4766     const SmallVectorImpl<FormatToken *> &Comments,
4767     const FormatToken *NextTok) {
4768   // Whether or not a line comment token continues a line is controlled by
4769   // the method continuesLineCommentSection, with the following caveat:
4770   //
4771   // Define a trail of Comments to be a nonempty proper postfix of Comments such
4772   // that each comment line from the trail is aligned with the next token, if
4773   // the next token exists. If a trail exists, the beginning of the maximal
4774   // trail is marked as a start of a new comment section.
4775   //
4776   // For example in this code:
4777   //
4778   // int a; // line about a
4779   //   // line 1 about b
4780   //   // line 2 about b
4781   //   int b;
4782   //
4783   // the two lines about b form a maximal trail, so there are two sections, the
4784   // first one consisting of the single comment "// line about a" and the
4785   // second one consisting of the next two comments.
4786   if (Comments.empty())
4787     return;
4788   bool ShouldPushCommentsInCurrentLine = true;
4789   bool HasTrailAlignedWithNextToken = false;
4790   unsigned StartOfTrailAlignedWithNextToken = 0;
4791   if (NextTok) {
4792     // We are skipping the first element intentionally.
4793     for (unsigned i = Comments.size() - 1; i > 0; --i) {
4794       if (Comments[i]->OriginalColumn == NextTok->OriginalColumn) {
4795         HasTrailAlignedWithNextToken = true;
4796         StartOfTrailAlignedWithNextToken = i;
4797       }
4798     }
4799   }
4800   for (unsigned i = 0, e = Comments.size(); i < e; ++i) {
4801     FormatToken *FormatTok = Comments[i];
4802     if (HasTrailAlignedWithNextToken && i == StartOfTrailAlignedWithNextToken) {
4803       FormatTok->ContinuesLineCommentSection = false;
4804     } else {
4805       FormatTok->ContinuesLineCommentSection =
4806           continuesLineCommentSection(*FormatTok, *Line, CommentPragmasRegex);
4807     }
4808     if (!FormatTok->ContinuesLineCommentSection &&
4809         (isOnNewLine(*FormatTok) || FormatTok->IsFirst)) {
4810       ShouldPushCommentsInCurrentLine = false;
4811     }
4812     if (ShouldPushCommentsInCurrentLine)
4813       pushToken(FormatTok);
4814     else
4815       CommentsBeforeNextToken.push_back(FormatTok);
4816   }
4817 }
4818 
4819 void UnwrappedLineParser::readToken(int LevelDifference) {
4820   SmallVector<FormatToken *, 1> Comments;
4821   bool PreviousWasComment = false;
4822   bool FirstNonCommentOnLine = false;
4823   do {
4824     FormatTok = Tokens->getNextToken();
4825     assert(FormatTok);
4826     while (FormatTok->isOneOf(TT_ConflictStart, TT_ConflictEnd,
4827                               TT_ConflictAlternative)) {
4828       if (FormatTok->is(TT_ConflictStart))
4829         conditionalCompilationStart(/*Unreachable=*/false);
4830       else if (FormatTok->is(TT_ConflictAlternative))
4831         conditionalCompilationAlternative();
4832       else if (FormatTok->is(TT_ConflictEnd))
4833         conditionalCompilationEnd();
4834       FormatTok = Tokens->getNextToken();
4835       FormatTok->MustBreakBefore = true;
4836       FormatTok->MustBreakBeforeFinalized = true;
4837     }
4838 
4839     auto IsFirstNonCommentOnLine = [](bool FirstNonCommentOnLine,
4840                                       const FormatToken &Tok,
4841                                       bool PreviousWasComment) {
4842       auto IsFirstOnLine = [](const FormatToken &Tok) {
4843         return Tok.HasUnescapedNewline || Tok.IsFirst;
4844       };
4845 
4846       // Consider preprocessor directives preceded by block comments as first
4847       // on line.
4848       if (PreviousWasComment)
4849         return FirstNonCommentOnLine || IsFirstOnLine(Tok);
4850       return IsFirstOnLine(Tok);
4851     };
4852 
4853     FirstNonCommentOnLine = IsFirstNonCommentOnLine(
4854         FirstNonCommentOnLine, *FormatTok, PreviousWasComment);
4855     PreviousWasComment = FormatTok->is(tok::comment);
4856 
4857     while (!Line->InPPDirective && FormatTok->is(tok::hash) &&
4858            (!Style.isVerilog() ||
4859             Keywords.isVerilogPPDirective(*Tokens->peekNextToken())) &&
4860            FirstNonCommentOnLine) {
4861       distributeComments(Comments, FormatTok);
4862       Comments.clear();
4863       // If there is an unfinished unwrapped line, we flush the preprocessor
4864       // directives only after that unwrapped line was finished later.
4865       bool SwitchToPreprocessorLines = !Line->Tokens.empty();
4866       ScopedLineState BlockState(*this, SwitchToPreprocessorLines);
4867       assert((LevelDifference >= 0 ||
4868               static_cast<unsigned>(-LevelDifference) <= Line->Level) &&
4869              "LevelDifference makes Line->Level negative");
4870       Line->Level += LevelDifference;
4871       // Comments stored before the preprocessor directive need to be output
4872       // before the preprocessor directive, at the same level as the
4873       // preprocessor directive, as we consider them to apply to the directive.
4874       if (Style.IndentPPDirectives == FormatStyle::PPDIS_BeforeHash &&
4875           PPBranchLevel > 0) {
4876         Line->Level += PPBranchLevel;
4877       }
4878       assert(Line->Level >= Line->UnbracedBodyLevel);
4879       Line->Level -= Line->UnbracedBodyLevel;
4880       flushComments(isOnNewLine(*FormatTok));
4881       parsePPDirective();
4882       PreviousWasComment = FormatTok->is(tok::comment);
4883       FirstNonCommentOnLine = IsFirstNonCommentOnLine(
4884           FirstNonCommentOnLine, *FormatTok, PreviousWasComment);
4885     }
4886 
4887     if (!PPStack.empty() && (PPStack.back().Kind == PP_Unreachable) &&
4888         !Line->InPPDirective) {
4889       continue;
4890     }
4891 
4892     if (FormatTok->is(tok::identifier) &&
4893         Macros.defined(FormatTok->TokenText) &&
4894         // FIXME: Allow expanding macros in preprocessor directives.
4895         !Line->InPPDirective) {
4896       FormatToken *ID = FormatTok;
4897       unsigned Position = Tokens->getPosition();
4898 
4899       // To correctly parse the code, we need to replace the tokens of the macro
4900       // call with its expansion.
4901       auto PreCall = std::move(Line);
4902       Line.reset(new UnwrappedLine);
4903       bool OldInExpansion = InExpansion;
4904       InExpansion = true;
4905       // We parse the macro call into a new line.
4906       auto Args = parseMacroCall();
4907       InExpansion = OldInExpansion;
4908       assert(Line->Tokens.front().Tok == ID);
4909       // And remember the unexpanded macro call tokens.
4910       auto UnexpandedLine = std::move(Line);
4911       // Reset to the old line.
4912       Line = std::move(PreCall);
4913 
4914       LLVM_DEBUG({
4915         llvm::dbgs() << "Macro call: " << ID->TokenText << "(";
4916         if (Args) {
4917           llvm::dbgs() << "(";
4918           for (const auto &Arg : Args.value())
4919             for (const auto &T : Arg)
4920               llvm::dbgs() << T->TokenText << " ";
4921           llvm::dbgs() << ")";
4922         }
4923         llvm::dbgs() << "\n";
4924       });
4925       if (Macros.objectLike(ID->TokenText) && Args &&
4926           !Macros.hasArity(ID->TokenText, Args->size())) {
4927         // The macro is either
4928         // - object-like, but we got argumnets, or
4929         // - overloaded to be both object-like and function-like, but none of
4930         //   the function-like arities match the number of arguments.
4931         // Thus, expand as object-like macro.
4932         LLVM_DEBUG(llvm::dbgs()
4933                    << "Macro \"" << ID->TokenText
4934                    << "\" not overloaded for arity " << Args->size()
4935                    << "or not function-like, using object-like overload.");
4936         Args.reset();
4937         UnexpandedLine->Tokens.resize(1);
4938         Tokens->setPosition(Position);
4939         nextToken();
4940         assert(!Args && Macros.objectLike(ID->TokenText));
4941       }
4942       if ((!Args && Macros.objectLike(ID->TokenText)) ||
4943           (Args && Macros.hasArity(ID->TokenText, Args->size()))) {
4944         // Next, we insert the expanded tokens in the token stream at the
4945         // current position, and continue parsing.
4946         Unexpanded[ID] = std::move(UnexpandedLine);
4947         SmallVector<FormatToken *, 8> Expansion =
4948             Macros.expand(ID, std::move(Args));
4949         if (!Expansion.empty())
4950           FormatTok = Tokens->insertTokens(Expansion);
4951 
4952         LLVM_DEBUG({
4953           llvm::dbgs() << "Expanded: ";
4954           for (const auto &T : Expansion)
4955             llvm::dbgs() << T->TokenText << " ";
4956           llvm::dbgs() << "\n";
4957         });
4958       } else {
4959         LLVM_DEBUG({
4960           llvm::dbgs() << "Did not expand macro \"" << ID->TokenText
4961                        << "\", because it was used ";
4962           if (Args)
4963             llvm::dbgs() << "with " << Args->size();
4964           else
4965             llvm::dbgs() << "without";
4966           llvm::dbgs() << " arguments, which doesn't match any definition.\n";
4967         });
4968         Tokens->setPosition(Position);
4969         FormatTok = ID;
4970       }
4971     }
4972 
4973     if (FormatTok->isNot(tok::comment)) {
4974       distributeComments(Comments, FormatTok);
4975       Comments.clear();
4976       return;
4977     }
4978 
4979     Comments.push_back(FormatTok);
4980   } while (!eof());
4981 
4982   distributeComments(Comments, nullptr);
4983   Comments.clear();
4984 }
4985 
4986 namespace {
4987 template <typename Iterator>
4988 void pushTokens(Iterator Begin, Iterator End,
4989                 llvm::SmallVectorImpl<FormatToken *> &Into) {
4990   for (auto I = Begin; I != End; ++I) {
4991     Into.push_back(I->Tok);
4992     for (const auto &Child : I->Children)
4993       pushTokens(Child.Tokens.begin(), Child.Tokens.end(), Into);
4994   }
4995 }
4996 } // namespace
4997 
4998 std::optional<llvm::SmallVector<llvm::SmallVector<FormatToken *, 8>, 1>>
4999 UnwrappedLineParser::parseMacroCall() {
5000   std::optional<llvm::SmallVector<llvm::SmallVector<FormatToken *, 8>, 1>> Args;
5001   assert(Line->Tokens.empty());
5002   nextToken();
5003   if (FormatTok->isNot(tok::l_paren))
5004     return Args;
5005   unsigned Position = Tokens->getPosition();
5006   FormatToken *Tok = FormatTok;
5007   nextToken();
5008   Args.emplace();
5009   auto ArgStart = std::prev(Line->Tokens.end());
5010 
5011   int Parens = 0;
5012   do {
5013     switch (FormatTok->Tok.getKind()) {
5014     case tok::l_paren:
5015       ++Parens;
5016       nextToken();
5017       break;
5018     case tok::r_paren: {
5019       if (Parens > 0) {
5020         --Parens;
5021         nextToken();
5022         break;
5023       }
5024       Args->push_back({});
5025       pushTokens(std::next(ArgStart), Line->Tokens.end(), Args->back());
5026       nextToken();
5027       return Args;
5028     }
5029     case tok::comma: {
5030       if (Parens > 0) {
5031         nextToken();
5032         break;
5033       }
5034       Args->push_back({});
5035       pushTokens(std::next(ArgStart), Line->Tokens.end(), Args->back());
5036       nextToken();
5037       ArgStart = std::prev(Line->Tokens.end());
5038       break;
5039     }
5040     default:
5041       nextToken();
5042       break;
5043     }
5044   } while (!eof());
5045   Line->Tokens.resize(1);
5046   Tokens->setPosition(Position);
5047   FormatTok = Tok;
5048   return {};
5049 }
5050 
5051 void UnwrappedLineParser::pushToken(FormatToken *Tok) {
5052   Line->Tokens.push_back(UnwrappedLineNode(Tok));
5053   if (MustBreakBeforeNextToken) {
5054     Line->Tokens.back().Tok->MustBreakBefore = true;
5055     Line->Tokens.back().Tok->MustBreakBeforeFinalized = true;
5056     MustBreakBeforeNextToken = false;
5057   }
5058 }
5059 
5060 } // end namespace format
5061 } // end namespace clang
5062